General Outline

General Outline
9 ULSI technologies: manufacturing
9 Radiation effects in devices and
technologies
9 Radiation Hardness-By-Design (HBD)
techniques for ASICs
9 130nm technology node for High
Energy Physics: vendor, properties,
radiation effects
IN2P3 school, May 2005
Federico Faccio - CERN
1
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
2
Foreword
9 What we will be talking about:
z
z
z
z
ASICs (application-specific ICs)
CMOS technologies only
Effects: TID, SEU, SEL
How to conceive ASICs able of surviving
to and functioning in a radiation
environment
IN2P3 school, May 2005
Federico Faccio - CERN
3
Is it a problem for me?
Which TID can be considered as “safe”?
Is there a dependence on design (logic vs analog, …)?
Is there variability with time (different results for different
production lots)?
9 Is there a technological dependence on the TID tolerance?
9
9
9
0.5μm
0.35μm
0.5μm
0.25μm
1.E -03
1.E -04
Leakage current [A]
1.E -05
1.E -06
1.E -07
1.E -08
1.E -09
1.E -10
1.E -11
1.E -12
0
50
100
150
200
250
T o ta l D o s e [k ra d (s iO 2 )]
IN2P3 school, May 2005
Federico Faccio - CERN
4
Risk management
9
9
9
9
As seen before, there are many variables influencing
the final radiation hardness of an IC
The “safer” the circuit is designed, the more
“expensive” it is (area, performance, complexity, …)
Risk reduction comes at some cost
In the following, a procedure to RELIABLY designing
radiation tolerant ASICs is presented. It reposes on
physics, not on specific process recipes, hence it
gives very predictable results.
IN2P3 school, May 2005
Federico Faccio - CERN
5
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
6
Radiation effects and tox scaling
Damage decreases with gate oxide thickness
Oxide trapped charge
Interface states
N.S. Saks et al., IEEE TNS, Dec. 1984 and Dec. 1986
IN2P3 school, May 2005
Federico Faccio - CERN
7
Tendency confirmed
9
Gate oxides in commercial CMOS technologies did follow the
curve drawn by Saks and co-workers!
Technology
node (μm)
DVth/Mrad(SiO2) [V/rad(SiO2)]
1.E+02
1.6
1.2
0.8
0.5
0.5 - A
0.5 - B
0.35
0.25 - A
0.25 - B
tox^2
1.E+01
1.E+00
1.E-01
1.E-02
1.E-03
1
10
100
tox (nm)
IN2P3 school, May 2005
Federico Faccio - CERN
8
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
9
Radiation hard CMOS processes
“Usual solution”: technology hardening
TID hardness is guaranteed
but
9 Low volume in fabs => yield can be low, and
unreliable radiation performance for large
quantities
9 Cost is very high
9 Very limited number of processes still available
today, and risk of unavailability in the long run
9 Analog performance often not very good
9
9
IN2P3 school, May 2005
Federico Faccio - CERN
10
Radiation-tolerant layout (ELT)
Leakage path
G
S
IN2P3 school, May 2005
G
D
D
Federico Faccio - CERN
S
11
Function of the guardring
+
N+ WELL CONTACT
VSS
VSS
VDD
OXIDE+
+ +
+
+
+
+
P+ GUARD
+ +
+
+
+
+
N+ SOURCE
N WELL
SUBSTRATE
IN2P3 school, May 2005
The heavily doped p+ guardring
“interrupts” possible conductive
paths (inverted p substrate), hence
preventing inter-device
leakage currents
Federico Faccio - CERN
12
Radiation tolerant approach
ΔVth ∝ toxn
+
ELT’s and
guard rings
=
TID
Radiation
Tolerance
IN
OUT
Example layout for an inverter
p+ guard ring
VSS
IN2P3 school, May 2005
metal
polysilicon
n+ diffusion
p+ diffusion
Federico Faccio - CERN
VDD
13
Effectiveness of ELTs
0.7 μm technology - W/L = 2000/1.5
1.E-02
1.E-03
1.E-04
1.E-05
ID [A]
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
Prerad
1.E-11
After 1 Mrad
1.E-12
After 1 Mrad (ELT)
1.E-13
-1
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
VG [V]
IN2P3 school, May 2005
Federico Faccio - CERN
14
Effectiveness of ELTs
0.25 μm technology - W/L = 30/0.4 - ELT
1.E-01
1.E-03
ID [A]
1.E-05
1.E-07
1.E-09
Prerad and
after 13 Mrad
1.E-11
1.E-13
-0.60 -0.22 0.16 0.54 0.92 1.30 1.68 2.06 2.44 2.82
V G [V]
IN2P3 school, May 2005
Federico Faccio - CERN
15
Field oxide leakage
FOXFET 14.4/2.6 without gate, with guardring
0.5 μm technology
ID [A]
1.E-10
1.9 Mrad
1.E-11
anneal
420 Krad
Prerad
1.E-12
0.0
0.5
1.0
1.5
2.0
2.5
3.0
VD [V]
IN2P3 school, May 2005
Federico Faccio - CERN
16
Advantages of this approach
9 Relies on physics (thickness of gate
oxide): not process-dependent
9 Allows for using state-of-the-art
technologies:
z
z
z
z
Low power
High performance
High throughput, high yield, short
turnaround times
Low cost
IN2P3 school, May 2005
Federico Faccio - CERN
17
Difficulties for this approach
9 Peculiar ELT behaviour
z
z
z
Modeling of ELT (size W/L?)
Limitation in aspect ratio
Asymmetry
9 Lack of commercial library for digital
design
9 Loss of density
9 Yield and reliability???
IN2P3 school, May 2005
Federico Faccio - CERN
18
Modeling of ELTs (1)
Source
c
d'
d
Drain
IN2P3 school, May 2005
L
2
Federico Faccio - CERN
+2
+
19
Modeling of ELTs (2)
d – d’
2
W
2α
1– α
= 4
+ 2K
+3
d’
L eff
Leff
ln
1.13•ln 1
α
d’ – 2αLeff
( )
Ldrawn
0.28
0.36
0.5
1
3
5
IN2P3 school, May 2005
Estimated
(W/L)eff
14.8
11.3
8.3
5.1
3
2.6
Extracted
(W/L)eff
15
11.2
8.3
5.2
3.2
2.6
1 shape only supported (size
of “c” fixed)
9 Custom routines and layers
integrated in design kit for
extraction/design
checking/computation
9
Federico Faccio - CERN
20
Limitation in the aspect ratio
Aspect ratio = W/L
15
14
Effective aspect ratio
13
12
11
10
9
8
7
6
5
4
3
2
0
1
2
3
4
5
6
7
8
9
10
Gate Length [[μm]
m]
IN2P3 school, May 2005
Federico Faccio - CERN
21
Asymmetry (1)
L = 0.28 μm
GDI = 11.9 μS
GDO = 9.6 μS
1.4E-04
1.2E-04
I drain [A ]
1.0E-04
Drain Inside
Drain Outside
8.0E-05
6.0E-05
4.0E-05
2.0E-05
0.0E+00
0.0
0.5
1.0
1.5
2.0
2.5
Vds [ V ]
IN2P3 school, May 2005
Federico Faccio - CERN
22
Asymmetry (2)
O u tp u t C o n d . [u S ]
100.0
ELT Drain Inside
ELT Drain Outside
Normal Transistor
10.0
L (μm) ΔG/GDI
0.28
0.36
0.5
1
3
5
1.0
0.1
19 %
23 %
33 %
53 %
70 %
75 %
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Gate Length [um ]
IN2P3 school, May 2005
Federico Faccio - CERN
23
Lack of commercial library (1)
Radiation tolerant
design :
z Use of enclosed
NMOS transistors
z Use of guard rings to
isolate all n+
diffusions at different
potentials (including
n-wells)
NAND3
IN2P3 school, May 2005
Federico Faccio - CERN
24
Lack of commercial library (2)
9
Example of core cells
IN2P3 school, May 2005
Federico Faccio - CERN
25
Lack of commercial library (3)
9
IN2P3 school, May 2005
Federico Faccio - CERN
List of
Library
Standard
Cells
26
Existing libraries
9 List is not exclusive:
z
CERN in 0.25μm
• “Daughter” libraries in PSI, LBL, Fermilab
z
z
IMEC (for ESA) in 0.18μm
Mission Research Corporation (for
Aerospace Corporation) in 0.18μm
IN2P3 school, May 2005
Federico Faccio - CERN
27
Loss of density
9
Radiation Tolerant
techniques introduce a
~70% layout area
overhead
Inverter
2-in NAND
2-in NOR
Static D-F/F
Static D-F/F SR
Standard
33.6 μm2
46.0 μm2
47.8 μm2
153.0 μm2
188.1 μm2
Rad-Tol
50.9 μm2
119.0 μm2
80.0 μm2
533.1 μm2
572.0 μm2
Penalty
34 %
61 %
41 %
71 %
75 %
0.8 μm
9
0.24 μm
Gate density is 8 times
larger when compared to
a 0.8μm technology
z
Example: ring oscillator
with 1280 inverters in 0.8
and 0.25mm technologies
(0.25 uses the CERN radtol
library)
a)
b)
Area:
IN2P3 school, May 2005
Federico Faccio - CERN
650,400 μm
82,350 μm
28
Yield and reliability?
9
9
9
Construction
analysis performed
on several chips
Circuits produced,
qualified and tested
in thousands (100
different designs!)
No concern on yield
or reliability found
yet…
IN2P3 school, May 2005
Federico Faccio - CERN
29
“Large” scale production
9
Relatively large number of project in production, for quantities
up to almost 600 wafers (200mm size). Total production in
excess of 3000 wafers.
Volume per project
700
500
400
300
Production summary
200
900
100
800
0
18
16
N of wafers produced
N of wafers produced
A PV
25
A tla s
pix
C MS
Pix
H PT
DC
T ra c
k e rS
e rv i c
e
Me d
A li ce ip ix
Pix e
ls
H AL
25
MG P
A
A DC
4 1 24
0
F e n ix
D iC a
Sy
L HC
B Pix
GO L
A NN
D tmr
o c -s
B e et
le
A li c e
TO F
P asc
Am
P re s
h o we
MP W r
11
C RT
A ll
MP W
15
MP W
14
700
14
N of projects in production
600
12
500
10
400
8
300
6
200
4
100
2
0
0
Y 1999 Y 2000 Y 2001 Y 2002 Y 2003 Y 2004 Y 2005 Y 2006
IN2P3 school, May 2005
Federico Faccio - CERN
30
N of projects in production
Volume (wafers)
600
ASICs examples
9
9
9
9
APV25: readout
of Silicon tracker
detector of the
CMS experiment
128 channels,
analog output
Alice Pixel1, readout of
silicon pixel detector of
the ALICE experiment
2.1cm2, 8000 analog
channels, 13M
transistors
IN2P3 school, May 2005
Federico Faccio - CERN
9
GOL, optical link
driver (serializer
(serializer +
laser driver) @ 1.6
Gbit/s
9
CCU, control chip for the
CMS tracker detector
(fully digital chip,
120kgates)
31
Other edgeless designs
9
Other “edgeless” transistor designs are possible !
G
G
from D.Mavis, MRC
S
D
Ringed Source
S
D
S
Ringed Interdigitated
G
S
D
Butterfly
IN2P3 school, May 2005
Federico Faccio - CERN
32
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
33
SEU and scaling: forecast
• VDD reduced
• Node C reduced
• New mechanisms
for SEU
P.E. Dodd et al., IEEE TNS, Dec. 1996
From the above, it looks like the SEU problem worsens with
scaling
IN2P3 school, May 2005
Federico Faccio - CERN
34
SEU and scaling: reality
All sources agree: DRAM sensitivity has been scaling down (cell
area scaling has outpaced the decrease in stored charge).
9 Picture somewhat less clear for SRAMs
9 P.Hazuka et al (work funded by Intel) developed a model to
predict SER scaling with Lg. The results suggest that the per-bit
sensitivity decreases –at least- linearly with Lg
9 Overall: FIT/MB decreases, but FIT/chip increases
9
9
Not only Vdd and node capacitance have to be taken into
account: sensitive area and charge collection efficiency are also
important and change with technology generation!
IN2P3 school, May 2005
Federico Faccio - CERN
35
SEL and scaling
9
9
9
Retrograde wells
Trench isolation
VDD reduced
VDD
contact
All these issues help in
preventing SEL, but
they might not be always
effective
VDD
source
n +
VSS
source
VSS
contact
n +
p +
p +
R2
n well
decrease
R1
p substrate
IN2P3 school, May 2005
R5
R6
R4
R3
decrease
Federico Faccio - CERN
36
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
37
Solutions: SEU
9
9
9
Technology level: epitaxial substrates, SOI,…
Cell design: SEU-tolerant FFs or memories
Redundancy
z
z
9
Triple Modular Redundancy (TMR): triplication and
voting
Encoding (EDAC)
Always to be considered at system level
IN2P3 school, May 2005
Federico Faccio - CERN
38
Cell design (1)
9
Increase the critical charge by
increasing the node capacitance:
z
Design larger transistors – also more driving
strength
IN2P3 school, May 2005
Federico Faccio - CERN
39
Cell design (2)
9
Increase the critical charge by increasing the
node capacitance:
Add “extra” capacitors
z
• Metal/metal to avoid loosing space
2
σ [cm /bit]
1.E-06
1.E-07
"standard" SR
"oversized" SR
"overloaded" SR
1.E-08
0
10
20
30
40
50
60
70
80
2
LET [MeVcm /mg]
IN2P3 school, May 2005
Federico Faccio - CERN
Upset rates in
proton environment:
- twofold decrease
for the “oversized”
- tenfold decrease
for the “overloaded”
40
Cell design (3)
DICE (Dual Interlock Cell)
9 Dual Interlock ensures SEU protection against hit on one node
9 Writing in the cell requires access to 2 nodes
9
0
IN2P3 school, May 2005
1
Federico Faccio - CERN
0
1
0
41
Cell design (4)
9
DICE cell weakness:
z
Recovery time needed after SEU
• Output glitch
• A rising edge of the clock during recovery time can
store wrong data in the following pipeline stage
z
z
Local clock buffers
Charge collection by multiple nodes is not
negligible! (in 90nm technology, just 10x SEU
rate improvement)
IN2P3 school, May 2005
Federico Faccio - CERN
42
Cell design (5)
9
Use special cell architectures
z
Temporal sampling with internal clock delays
(after Mavis) effective against digital SET
• Transient can only be
captured by 1 latch
• Sensitive to
transients on clock line
• Many variations to
this concept exist (one
can delay data instead
of clock, for instance)
IN
D Q
DFF
CLOCK
OUT
ΔT
2ΔT
D Q
DFF
D Q
DFF
Temporal Sampling
IN2P3 school, May 2005
Federico Faccio - CERN
MAJ
Asynchronous Voting
43
Solutions: SEU
9
9
9
Technology level: epitaxial substrates, SOI,…
Cell design: SEU-tolerant FFs or memories
Redundancy
z
z
9
Triple Modular Redundancy (TMR): triplication and
voting
Encoding (EDAC)
Always to be considered at system level
IN2P3 school, May 2005
Federico Faccio - CERN
44
Redundancy: TMR (1)
9
9
9
9
Triplication with 1
voter
The state machine is instantiated
3 times, with 1 voter
An SEU can corrupt the output of
one of the blocks, but majority
voting restores the correct state
An error in the voter instead
corrupts the state!
IN2P3 school, May 2005
Federico Faccio - CERN
Combinatorial
Logic
Register
Combinatorial
Logic
Register
Combinatorial
Logic
Register
Majority
Voter
45
Redundancy: TMR (2)
9
Triplication with 3 voters
9
The state machine is instantiated 3
times, with 3 voters
An SEU can corrupt the output of one
of the blocks, but majority voting
restores the correct state
An error in one of the voters is also
restored
9
9
9
Combinatorial
Logic
Register
Majority
Voter
Combinatorial
Logic
Register
Majority
Voter
Combinatorial
Logic
Register
Majority
Voter
This seems a lot of additional
transistors, but sometimes the designer
can find more “compact solutions”. This
is shown in the following example.
IN2P3 school, May 2005
Federico Faccio - CERN
46
Example of TMR in registers (1)
A majority voter can necessitate a
large number of transistors. Since 3
voters are required to protect the data
from errors in the voter, the overall
area penalty can be very large!
9 One compact solution for the voter is in
the schematic below:
9
9
9
9
9
IN2P3 school, May 2005
Federico Faccio - CERN
Combinatorial
Logic
Register
Majority
Voter
Combinatorial
Logic
Register
Majority
Voter
Combinatorial
Logic
Register
Majority
Voter
Very compact layout
14 transistors (better than
XOR+MUX style)
Fast
z
2 logic levels only
z
Possible to save the
inverter (2 transistors less)
Still 3 voters are needed, for a
minimum of 36 transistors
47
Example of TMR in registers (2)
9
9
To save some transistors, it is possible to merge the voter in the
register, and make a single cell
Each “super cell” will then have 3 inputs and 1 output with merged
connections between them
Register
Register
Register
in1
Majority
Voter
Majority
Voter
B
C
in2
D0 Q
D1
D2
A
out1
A
A
C
B
D0 Q
D1
D2
out2
B
Majority
Voter
in3
D0 Q
D1
D2
A
B
D0 Q
D1
D2
C out3
C
ck
IN2P3 school, May 2005
Federico Faccio - CERN
48
Example of TMR in registers (3)
Merging of the voter can be done at different levels
The example below, for a dynamic FF, illustrates it. The first part of the FF cell is
shown in the left picture, with its timing diagram.
A first level of merging is achieved by integrating the voter in the input inverter,
as shown in the right picture. This does not really save transistors with respect
to a separate voter
9
9
9
Clock =
in1
D0 Q
D1
D2
B
C
in1
a
C
B
B
out1
A
b
in
A
in1
C
b
Clock
a
a
in
b
precharge
IN2P3 school, May 2005
a
a
in
Federico Faccio - CERN
49
Example of TMR in registers (4)
9
A “deeper” merging, where the voter is moved inside
the FF, leads to considerable area saving
in1
D0 Q
A
aB D1 a
aC D2
out1
aA
A
b
aB
aA
aC
aA
aC
aB
in
aA
IN2P3 school, May 2005
Federico Faccio - CERN
This integration of the
voter in the cell “only”
costs 5 additional
transistors! For 3
“super cells”, the
penalty for the voter
decreases from 36 to
15 transistors!
50
Solutions: SEU
9
9
9
Technology level: epitaxial substrates, SOI,…
Cell design: SEU-tolerant FFs or memories
Redundancy
z
z
9
Triple Modular Redundancy (TMR): triplication and
voting
Encoding (EDAC)
Always to be considered at system level
IN2P3 school, May 2005
Federico Faccio - CERN
51
Redundancy: encoding (1)
9
Adding redundant information (bits) and
encoding-decoding
z
z
z
Used for data transmission and for memories
Requires complex encoding-decoding logic
Several different codes can be used
(Hamming, Reed-Solomon, BCH, etc.)
Encoder
Decoder
Memory array (each word with redundant bits)
IN2P3 school, May 2005
Federico Faccio - CERN
52
Redundancy: encoding (2)
9
Example: Hamming encoding (1950)
k = number of message bits
q = number of check (parity) bits
n = number of word bits
Minimum distance between words = 3
⇒All valid words in the code differ AT LEAST by 3 bits
⇒It can be used for single error correction, double error detection
n = k+q
n ≤ 2q-1
30
30 28
26
24
22
20
18
16
k( q)
14
12
10
8
6
4
1 2
0
2
2
IN2P3 school, May 2005
Federico Faccio - CERN
3
4
q
5
30
28 30
26
24
22
20
18
16
n ( q)
14
12
10
8
6
4
2 3
0
5
53
Redundancy: encoding (3)
9
Example: Hamming encoding (1950)
Example of encoding for
k=8
q=4
n = 12
Encode in a way requiring “as easy as possible” encoding/decoding logic
1. Check bits in powers of 2 positions in the word (position 1,2,4,8)
2. Other word bits are the message bits
3. Each check bit computes the parity for some of the word bits:
Position 1: check 1 bit, skip 1 bit, etc. (bits 1,3,5,7,9,11)
Position 2: check 2 bits, skip 2 bits, etc. (bits 2,3,6,7,10,11)
Position 4: check 4 bits, skip 4 bits, etc. (bits 4,5,6,7,12)
Position 8: check 8 bits, skip 8 bits, etc. (bits 8,9,10,11,12)
Word to encode: 10101010
word
position
IN2P3 school, May 2005
. 1 1
. 1
1
. 0 1 0 0
. 1 0 1 0
1 2 3 4 5 6 7 8 9 10 11 12
Federico Faccio - CERN
54
Redundancy: encoding (4)
9
Example: Hamming encoding (1950)
Word to encode: 10101010
Encoded word 1 1 1 1 0 1 0 0 1 0 1 0
position 1 2 3 4 5 6 7 8 9 10 11 12
SEU changes to 1 1 1 1 0 0 0 0 1 0 1 0
Check the parity bits in the received word:
Position 1: OK
Position 2: wrong
Position 4: wrong
Position 8: OK
The position of wrong bit is the sum of the wrong positions, that is
Position 6!
IN2P3 school, May 2005
Federico Faccio - CERN
55
Outline
9
9
Foreword: CMOS technologies
TID: Total Ionizing Dose
z
z
z
9
Effects (reminder)
Solutions
Trends in state-of-the-art technologies
SEEs: Single Event Effects
z
z
z
Effects (reminder)
SEEs and scaling
Solutions
• SEU
• SEL
9
Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
56
Reducing SEL sensitivity
The best solution is to decrease the gain of the parasitic pnpn structure.
Technological and layout solution can help in that respect:
Technological
=> use of epitaxial substrates and retrograde wells
=> use of trench instead of junction isolation
Layout
=> increase the distance between complementary devices
=> use guardrings
=> use lots of substrate and well contacts
VDD
contact
VDD
source
n +
VSS
guardring
VSS
source
VSS
contact
p +
n +
p +
p +
R2
n well
R4
R6
R1
p substrate
IN2P3 school, May 2005
R5
R3
Federico Faccio - CERN
57
Hardness-By-Design (HBD) techniques
Outline
9 Foreword
9 TID: Total Ionizing Dose
z
z
TID and scaling
Solutions
9 SEEs: Single Event Effects
z
z
SEEs and scaling
Solutions
9 Summary & Conclusion
IN2P3 school, May 2005
Federico Faccio - CERN
58
Summary & Conclusion
9
9
9
9
In state-of-the-art commercial-grade CMOS
technologies, reliable radiation tolerance can be
achieved with HBD techniques
If “natural” TID is insufficient, the use of ELTs and
guardrings can “upgrade” it – large effort needed
mainly for digital design
SEU rates can be decreased with proper
engineering (study of environment, introduction of
tolerant cells, redundancy, encoding, etc.) and
should always be considered at the system level
SEL – already unlikely – can be further controlled by
careful layout
IN2P3 school, May 2005
Federico Faccio - CERN
59
Further reading
9
On Enclosed Transistor Layout:
z
z
z
9
Ph.D. thesis describing results in 0.25μ
0.25μm technology:
technology:
G.Anelli,
G.Anelli, “Conception et caractérisation de circuits intégrés résistants aux radiations pour les détecteurs de particules du
LHC en technologies CMOS submicroniques profondes“, Ph.D. Thesis at the Politechnic School of Grenoble (INPG),
France, December 2000, availeble on the web at the URL: http://rd49.web.cern.ch/RD49/RD49Docs/anelli/these.html
http://rd49.web.cern.ch/RD49/RD49Docs/anelli/these.html
Paper containing all references on the work done at CERN on this subject: F.Faccio,
F.Faccio, “Radiation Issues in the new
generation of high energy physics experiments”, Int. Journal of High Speed Electronics and Systems, Vol.14, No.2 (2004)
379379-399
On SEUSEU-tolerant Cells:
z
Increased capacitance:
•
•
•
z
Special SEUSEU-tolerant cells:
•
•
•
•
•
9
F.Faccio et al., “Single Event Effects in Static and Dynamic Registers in
in a 0.25mm CMOS Technology”, IEEE Trans. Nucl.
Nucl.
Science, Vol.46, No.6, pp.1434pp.1434-1439, December 1999
F.Faccio et al., “SEU effects in registers and in a DualDual-Ported Static RAM designed in a 0.25mm CMOS technology for
applications in the LHC”, in the proceedings of the Fifth Workshop
Workshop on Electronics for LHC Experiments, Snowmass,
September 2020-24, 1999, pp.571pp.571-575 (CERN 9999-09, CERN/LHCC/99CERN/LHCC/99-33, 29 October 1999)
P.Roche, F.Jacquet, C.Caillat, J.P.Schoellkopf, "An Alpha Immune and Ultra Low Neutron SER High Density SRAM",
proceedings of IRPS 2004, pp671-672, April 2004
R.Velazco et al., “2 CMOS Memory Cells Suitable for the Design of SEUSEU-Tolerant VLSI Circuits”, IEEE Trans. Nucl.
Nucl. Science,
Vol.41, No.6, p.2229, December 1994
T.Calin et al., “Upset Hardened Memory Design for Submicron CMOS Technology”,
Technology”, IEEE Trans. Nucl.
Nucl. Science, Vol.43, No.6,
p.2874, December 1996
M.N.Liu et al., “Low power SEU immune CMOS memory circuits”, IEEE Trans.
Trans. Nucl.
Nucl. Science, Vol.39, No.6, p.1679,
December 1992
J.Canaris,
J.Canaris, S.Whitaker,
S.Whitaker, “Circuit techniques for the radiation environment of Space”, IEEE
IEEE 1995 Custom Integrated Circuits
Conference, p.77
P.Eaton,
P.Eaton, D.Mavis et al., “Single Event Transient Pulsewidth Measurements Using a Variable Temporal Latch Technique”,
IEEE Trans. Nucl.
Nucl. Science, Vol.51, no.6, p.3365, December 2004
On TMR and encoding:
z
Paper comparing techniques and containing references, to be used as a starting research point: S.Niranjan,
S.Niranjan, J.F.Frenzel,
J.F.Frenzel,
“A comparison of FaultFault-Tolerant State Machine Architectures for SpaceSpace-Borne Electronics”, IEEE Trans. On Reliability,
Vol.45, No1, p.109, March 1996
IN2P3 school, May 2005
Federico Faccio - CERN
60
General Outline
9 ULSI technologies: manufacturing
9 Radiation effects in devices and
technologies
9 Radiation Hardness-By-Design (HBD)
techniques for ASICs
9 130nm technology node for High
Energy Physics: vendor, properties,
radiation effects
IN2P3 school, May 2005
Federico Faccio - CERN
61
130nm technology node for HEP
Outline
9 The 130nm technology node
9 Radiation effects in samples from
different Foundries
9 CERN Foundry Service in 130nm
z
z
z
Specific features
Radiation effects
Design Kits
IN2P3 school, May 2005
Federico Faccio - CERN
62
The 130nm technology node
In production since 2000-2001
Still available via “brokers” (Europractice, MOSIS, CMP), but
not all versions
9 General common features:
9
9
z
z
z
z
z
z
z
z
z
z
“still” available in 200mm wafers (8 inches)
8 levels of copper metallization
Logic and analog/RF versions with different devices available
2 oxide thicknesses (core 2-2.2nm and I/O for operation to 2.5V),
thin oxides that should be radiation tolerant
Wide range of transistors (different dopings)
Effective minimum L 0.08-0.09 um
Integration achievable of the order of 200kgates/mm2
Very wide range of IP blocks generally available from either the
Foundry or third parties
Large number of masks, and expensive => NRE between 2.5 and
5 times more expensive than “our” 250nm
Projected availability to around 2018
IN2P3 school, May 2005
Federico Faccio - CERN
63
130nm technology node for HEP
Outline
9 The 130nm technology node
9 Radiation effects in samples from
different Foundries
• Core transistors (ELT, linear layout)
• I/O transistors (ELT, linear layout)
9 CERN Foundry Service in 130nm
z
z
z
Specific features
Radiation effects
Design Kits
IN2P3 school, May 2005
Federico Faccio - CERN
64
Aim and experimental details
9
9
9
9
9
9
AIM: Understanding the natural radiation
tolerance of the 130nm node. Testing samples
from different Foundries allows to check if
observed results are common properties or
specific of just “one” particular process
Samples from 3 commercial 130nm CMOS
processes have been tested
Some are Process Monitoring Devices (PMD)
from foundry, some custom-designed test ICs
NMOS and PMOS transistors, core and I/O
devices (different oxide thickness), FOXFETs
Testing done at probe station – no bonding
required (transistors are extremely fragile and
gate oxide can be damaged by bonding)
Irradiation with X-rays at CERN up to 100200Mrad, under worst case static bias
The CERN X-ray facility
Custom-developed
test IC
IN2P3 school, May 2005
Federico Faccio - CERN
65
Core NMOS transistors, enclosed (ELT)
Example: Foundry A
9
1.00E-03
1.00E-04
1.00E-05
1.00E-06
Id (A)
The radiation hardness
of the gate oxide is
such that practically no
effect is observed –
verified for 2 foundries
(A up to 140Mrad, B up
to 30Mrad). Id=f(Vgs)
curves drop on top of
each other after any
TID step (see figures)
ELT transistors are
suitable for SHLC
applications: are they
really needed?
3Mrad
1.00E-08
136Mrad
2d HT ann
1.00E-10
1.00E-11
1.00E-12
-0.2
0.3
0.8
1.3
Vg (V)
PMOS ELT min/012
1.E-03
1.E-04
1.E-05
1.E-06
1.E-07
pre-rad
3 Mrd
40 Mrd
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
-0.2
IN2P3 school, May 2005
Pre-rad
1.00E-07
1.00E-09
Id (A)
9
NMOS ELT min/0.12
Federico Faccio - CERN
0.3
0.8
Vg (V)
1.3
66
Core NMOS transistors, linear layout (1)
Wide transistors
(W > 1μm):
9
When the transistor is off or
in the weak inversion regime:
• Leakage current appears
(for all transistor sizes)
• Weak inversion curve is
distorted
z
When the transistor is in
strong inversion:
• No significant change in
characteristics
IN2P3 school, May 2005
1.E-02
1.E-03
1.E-04
Id (A)
z
Foundry A, 2/0.12
1.E-05
1.E-06
Pre-rad
3Mrad
136Mrad
2d HT ann
1.E-07
1.E-08
1.E-09
1.E-10
-0.5
Federico Faccio - CERN
0.0
0.5
1.0
1.5
2.0
Vg (V)
67
Core NMOS transistors, linear layout (2)
Narrow transistors
(W < 0.8μm):
9
When the transistor is off or
in the weak inversion regime:
• Leakage current appears
(for all transistor sizes)
• Weak inversion curve is
distorted
z
z
Additionally, a Vth shift
(decrease) appears for
narrow channel transistors
The narrower the transistor,
the larger the Vth shift. This
effect has been called
Radiation-Induced Narrow
Channel Effect (RINCE)
IN2P3 school, May 2005
1.E-03
1.E-04
1.E-05
1.E-06
Id (A)
z
Foundry A, 0.16/0.12
1.E-07
Pre-rad
3Mrad
136Mrad
2d HT ann
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
-0.5
Federico Faccio - CERN
0.0
0.5
1.0
1.5
2.0
Vg (V)
68
Core NMOS transistors, linear layout (3)
Let’s observe the effect on the
leakage current
z
z
z
Peak in leakage at a TID of 15Mrad (A and B), similar for all
transistor sizes
Peaking dependent on dose
rate and temperature, difficult
to estimate in real environment
Different in the 3 technologies
0.16/0.12
0.32/0.12
0.48/0.12
0.8/0.12
2/0.12
10/1
10/10
ELT
Foundry A
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E+05
pre-rad
1.E+06
1.E+07
1.E-06
1.E-07
1.E-07
1.E-08
N_10_013
1.E-09
N_10_10
1.E-08
N_10_012
N_088_012
1.E-09
N_04_013
N_053_012
N_024_013
N_018_013
1.E-10
N_028_012
1.E-10
N_014_013
pre-rad
1.E+09
annealing
1.E-05
Foundry
C
1.E-06
1.E-11
1.E+04
1.E+08
TID (rad)
Ileak (A)
Ileak (A)
1.E-05
Foundry
B
1.E-05
Ileak (A)
9
1.E+05
IN2P3 school, May 2005
1.E+06
1.E+07
TID (rd)
1.E+08
N_016_012
1.E+09
1.E-11
1.E+04
pre-rad
Federico Faccio - CERN
1.E+05
1.E+06
1.E+07
1.E+08
annealing
1.E+09
TID (rd)
69
Core NMOS transistors, linear layout (4)
Let’s observe the effect on the
threshold voltage
z
Peak in Vth shift at a TID of 15Mrad (A and C)
z
The narrower the transistor, the
larger the Vth shift (RINCE)
z
Peaking dependent on dose rate
and temperature, difficult to
estimate in real environment
z
Sample from Foundry B different!
Foundry
A
0.000
-0.020
-0.040
-0.060
-0.120
-0.140
-0.160
1.E+05
-0.040
-0.060
-0.060
8 Vth (V)
-0.040
-0.080
-0.140
1.E+05
1.E+07
annealing
1.E+09
1.E+08
Foundry C
0.000
-0.020
N_10_10
-0.080
N_10_012
N_10_013
N_04_013
N_024_013
N_018_013
N_014_013
-0.120
1.E+06
TID (rad)
-0.020
-0.100
016_012
032_012
048_012
08_012
2_012
10_1
10_10
ELT
-0.080
-0.100
Foundry B
0.000
⎝ Vth (V)
0.020
Vth (V)
9
N_088_012
-0.100
N_053_012
N_028_012
-0.120
N_016_012
-0.140
1.E+06
1.E+07
1.E+08
1.E+09
1.E+05
IN2P3 school, May 2005
1.E+06
1.E+07
1.E+08
1.E+09
TID (rd)
TID (rd)
Federico Faccio - CERN
70
Radiation-induced edge effects - NMOS
• The peak at 1-5Mrd in the leakage
E field lines
and Vth can be explained by the
different evolution of charge trapped
Polysilicon gate
in the bulk of the lateral STI
++ ++
++ ++ -(positive) and in the interface states
STI ++ at the interface between STI oxide
- ++ STI
and silicon (negative)
• At the beginning of irradiation,
Oxide trapped
charge trapped in the oxide
Interface
Depletion
region
charge
dominates. With further TID,
states
interface states eventually dominate.
ID
The relative contribution will depend
Main transistor
on T, dose rate, etc.
• The narrower the transistor, the
Lateral
more important the influence of the
parasitic
charge trapped in the lateral STI and
transistor
interface on the full depletion region
=> the more relevant the radiation
effects => RINCE
0
VGS
IN2P3 school, May 2005
Federico Faccio - CERN
71
Core PMOS transistors, linear layout (1)
9
1.E-05
1.E-06
1.E-07
1.E-08
Pre-rad
3Mrad
30Mrad
136Mrad
1dHT
1.E-09
1.E-10
1.E-11
1.E-12
-0.5
0.0
0.5
1.0
1.5
2.0
Vg (V)
1.E-04
Foundry
C, 0.28/0.12
1.E-05
1.E-05
1.E-06
1.E-06
1.E-07
1.E-07
1.E-08
Id (A)
Id (A)
1.E-04
Foundry
B, 0.14/0.13
1.E-04
Foundry
A, 0.16/0.12
Id (A)
No change in the weak inversion
regime, no leakage
An apparent Vth shift (decrease)
for narrow channel transistors,
more visible in samples from
Foundry A
z
The narrower the transistor, the
larger the Vth shift (RINCE)
9
Pre-rad
1.E-08
pre-rad
22.5 Mrd
1.E-09
26 Mrd
1.E-09
89.5 Mrd
100 Mrd
1.E-10
1.E-10
1.E-11
1.E-11
1.E-12
1.E-12
-0.5
0.0
0.5
1.0
1.5
2.0
-0.5
IN2P3 school, May 2005
0.0
0.5
1.0
1.5
2.0
Vg (V)
Vg (V)
Federico Faccio - CERN
72
Radiation-induced edge effects - PMOS
• IN PMOS, the charge trapped in
E field lines
both oxide traps and interface states
Polysilicon gate
are positive. Therefore, the
threshold voltage of the lateral
++ ++
++ ++ -+
+
- ++
“parasitic” increases steadily and no
STI ++ +
+
leakage or deformation of the
STI
Id=f(Vgs) curve in weak inversion is
observed.
Oxide trapped
• Nevertheless, in narrow transistors
Depletion region Interface
charge
the charge trapped laterally can
states
influence the electric field in the
ID
whole depletion region, inducing and
Main transistor
apparent Vth shift.
•The narrower the transistor, the
Lateral
more important the influence of the
parasitic
charge trapped in the lateral STI and
transistor
interface on the full depletion region
=> the more relevant the radiation
effects => RINCE
0
VGS
IN2P3 school, May 2005
Federico Faccio - CERN
73
I/O transistors, ELT
9
Non-negligible effect
on the Vth, especially
for PFETs
Shift larger than what
observed in 250nm
node in the past…
1.E-02
NMOS
min/0.26
1.E-03
Id (A)
9
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
1.E-13
-0.5
pre-rad
3 Mrd
57 Mrd
190 Mrd
0.0
0.5
1.0
1.5
Vg (V)
2.0
2.5
3.0
1.E-03
Id (A)
PMOS
min/0.26
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
-0.5
pre-rad
1 Mrd
3 Mrd
40 Mrd
annealing
0.0
0.5
1.0
1.5
2.0
Vg (V)
IN2P3 school, May 2005
Federico Faccio - CERN
74
I/O transistors, linear layout
z
z
9
Peak degradation at 15Mrd observed as well
RINCE observed as well
Results different for
different Foundries, but
for all leakage in
NMOS starts already at
50-100krad
1.E-03
Foundry
A, NMOS 0.36/0.24
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
1.E-13
pre-rad
300 krd
1 Mrd
27 Mrd
97 Mrd
190 Mrd
-0.5
0.0
0.5
1.E-03
1.0
1.5
Vgs (V)
2.0
2.5
3.0
Foundry A, PMOS 2/0.24
1.E-04
1.E-05
1.E-06
Ids (A)
Same qualitative
effects as for core
transistors, but larger
impact on transistor
parameters
Ids (A)
9
1.E-07
1.E-08
1.E-09
pre-rad
1.E-10
1 Mrd
5 Mrd
1.E-11
25 Mrd
1.E-12
65 Mrd
1.E-13
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
Vgs (V)
IN2P3 school, May 2005
Federico Faccio - CERN
75
130nm technology node for HEP
Outline
9 The 130nm technology node
9 Radiation effects in samples from
different Foundries
9 CERN Foundry Service in 130nm
z
z
z
Specific features
Radiation effects
Design Kits
IN2P3 school, May 2005
Federico Faccio - CERN
76
Selected Foundry Services supplier
9
9
9
9
Competitive tendering has taken place in 2006
(CERN rules for “frame contracts”)
One Foundry has been selected to supply the
serviceContract is for 130nm (and beyond) CMOS
and BiCMOS
CERN will organize MPWs and engineering runs as
already provided in the 250nm node
In the absence of sufficient demand, MPW service is
now accessible via a broker. CERN organizes
“grouped” access via the broker, since there is a
minimum area to be purchased (10mm2). Grouping
allows users to prototype small chips for reasonable
charges
IN2P3 school, May 2005
Federico Faccio - CERN
77
130nm technology node for HEP
Outline
9 The 130nm technology node
9 Radiation effects in samples from
different Foundries
9 CERN Foundry Service in 130nm
z
z
z
Specific features
Radiation effects
Design Kits
IN2P3 school, May 2005
Federico Faccio - CERN
78
Natural radiation tolerance and its stability
9
9
9
9
The use of ELTs and guardring renders the ASICs
tolerant to TID safely and reliably – but the cost is
performance (power, area) and effort (custom library)
Can we rely on the “natural” tolerance of the
technology, such that “standard” layout can satisfy
our TID tolerance requirements?
The question is complex and answers will be
elaborated in the following few slides…
NB: Radiation tests have been performed on this
technology over time. Available hardware from Q12002 to Q1-2006 (4 samples at different times). In
this time frame, the process went through final
development and then production phase, possibly
with significant changes. Nevertheless, radiation
performance is very comparable!
IN2P3 school, May 2005
Federico Faccio - CERN
79
Core NMOS transistors
REMINDER of main effects:
Wide transistor (W>0.8um)
Narrow transistor (W>0.8um)
1.E-03
1.E-02
1.E-04
1.E-05
1.E-04
1.E-06
1.E-05
1.E-07
1.E-06
Pre-rad
3Mrad
136Mrad
2d HT ann
1.E-07
1.E-08
1.E-09
1.E-10
-0.5
Id (A)
Id (A)
1.E-03
Pre-rad
3Mrad
136Mrad
2d HT ann
1.E-08
1.E-09
1.E-10
1.E-11
1.E-12
0.0
0.5
1.0
1.5
-0.5
2.0
0.0
0.5
1.0
1.5
2.0
Vg (V)
Vg (V)
Leakage and “distortion” of weak
inversion slope => all sizes
Influence on “main” transistor
also in strong inversion =>
narrow transistors (RINCE)
There is NO sharp frontier between “narrow” and “wide”! RINCEs will
be increasingly pronounced for narrower transistors…
IN2P3 school, May 2005
Federico Faccio - CERN
80
Core PMOS transistors
REMINDER of main effects:
NO leakage or distortion
of weak inversion curve
1.E-04
1.E-05
1.E-06
Id (A)
1.E-07
1.E-08
Pre-rad
3Mrad
30Mrad
136Mrad
1dHT
1.E-09
1.E-10
1.E-11
1.E-12
-0.5
0.0
0.5
1.0
1.5
2.0
Vg (V)
Slight Vth shift for the narrowest
transistors only (0.16-0.3um)
IN2P3 school, May 2005
Federico Faccio - CERN
81
I/O transistors
Results qualitatively similar to Core FETs, but the
impact of damage is more important
0.200
1.E-05
0.100
1.E-06
0.000
1.E-07
10/1
-0.100
1.E-08
ELT
-0.200
0.36/0.24
0.8/0.24
2/0.24
10/1
10/10
ELT
-0.300
-0.400
-0.500
1.E+05
1.E+06
1.E+07
1.E+08
TID (rd)
1.E+09
annealing
Ileak (A)
NMOS
DVth (V)
9
0.36/0.24
0.8/0.24
2/0.24
10/10
1.E-09
1.E-10
1.E-11
1.E-12
1.E+04
1.E+05
1.E+06
1.E+07
1.E+08
annealing
1.E+09
TID (rd)
0 .5 0 0
0 .4 5 0
0 .4 0 0
PMOS
DV th (V
0 .3 5 0
0 .3 6 /0 .2 4
0 .8 /0 .2 4
Notice that the radiation effects
are important ALSO for ELTs
(also for PMOS, not shown
here)
2 /0 .2 4
1 0 /1
1 0 /1 0
0 .3 0 0
0 .2 5 0
0 .2 0 0
0 .1 5 0
0 .1 0 0
0 .0 5 0
0 .0 0 0
1 .E + 0 5
1 .E + 0 6
1 .E + 0 7
1 .E + 0 8
a n n1e.E
a lin
+ 0g9
TID (rd )
IN2P3 school, May 2005
Federico Faccio - CERN
82
Radiation-stability of Core NMOS FETs
Exactly the same “test structure” was integrated in 2 runs (about 16 months apart). Results are compared
here for the leakage current and Vth shift. They are well comparable (charts use the same scales)
1.E-0 6
0.16/0.12
0.32/0.12
0.48/0.12
0.8/0.12
2/0.12
10/1
10/10
E LT
1.E-07
1.E-0 8
1.E-08
Ileak (A)
Ileak (A)
1.E-0 7
1.E-06
1.E-0 9
1.E-09
016_012
032_012
1.E-1 0
048_012
1.E-10
064_012
08_012
1.E-1 1
1.E+0 4
pre-rad
2_012
1.E+05
1 .E+0 6
1 .E+0 7
1 .E+0 8
1.E-11
1.E+04
1 .E+0 9
annealing
10_1
1.E+05
1.E+06
TID (rd )
1.E+07
1.E+08
TID (rad)
Run “CuTe2”
10_10
1.E+09
ELT_min_0.12
Run “CuTe3”
0.020
0.020
0.000
0.000
-0.020
-0.020
-0.040
016_012
032_012
048_012
08_012
2_012
10_1
10_10
ELT
-0.080
-0.100
-0.120
-0.140
-0.160
1.E+05
1.E+06
1.E+07
1.E+08
annealing
1.E+09
Vth (V)
DVth (V)
-0.040
-0.060
-0.060
-0.080
-0.100
-0.120
-0.140
-0.160
1.E+05
1.E+06
1.E+07
1.E+08
1.E+09
TID (rad)
TID (rd)
IN2P3 school, May 2005
016_012
032_012
048_012
064_012
08_012
2_012
10_1
10_10
ELT_min_0.12
Federico Faccio - CERN
83
Radiation-stability of Core PMOS FETs
Exactly the same “test structure” was integrated in 2 runs (about 16 months apart). Results are compared
here for the Vth shift. They are well comparable (charts use the same scales)
0.100
0.100
016_012
0.080
016_012
08_012
048_012
0.080
2_012
08_012
10_1
2_012
10_10
10_1
0.060
10_10
0.040
° Vth (V)
DVth (V)
0.060
0.020
0.020
0.000
-0.020
1.E+05
0.040
0.000
1.E+06
1.E+07
1.E+08
annealing
1.E+09
-0.020
1.00E+05
TID (rd)
1.00E+06
1.00E+07
1.00E+08
1.00E+09
TID (rd)
Run “CuTe2”
Run “CuTe3”
Results were very comparable for all other devices measured (I/O transistors,
resistors, FOXFETs). CONCLUSION: the radiation tolerance does not seem
to vary significantly with time, and to be sensitive to small process changes or
fluctuations in the processing procedure.
IN2P3 school, May 2005
Federico Faccio - CERN
84
Ringed layout
Alternative shapes for “hardness by
design” based on ringed layout are not
very attractive anymore in the 130nm
technology node.
Influence of charge trapped in the STI
can influence the inversion condition
of the p- diffusion under the ring, and
leakage can still appear (unless the
overlap of gate on p- is very large)
G
D
S
G
S
S
D
D
D
Butterfly
Enclosed Layout Transistor (ELT)
1.E-03
L2
1.E-04
G
++++++++++++
----------------
1.E-05
1.E-06
pId (A)
S
S
Ringed Interdigitated
G
STI
++++++++++++
---------------p-
D
S
Ringed Source
STI
G
L1
G
S
D
1.E-07
1.E-08
pre-rad
300 krd
3 Mrd
40 Mrd
1.E-09
1.E-10
1.E-11
1.E-12
-0.5
0.0
0.5
1.0
1.5
2.0
Vg (V)
If overlap region L gets small (like L2),
L2), inversion in the pp- under the poly
can occur under the influence of the electric field induced by the
the charge
trapped in the STI (indicated with + in the drawings). Therefore Leakage
IN2P3 school, May 2005
Federico Faccio - CERN
appears
85
Inter-device leakage
Measurements based mainly on FOXFETs (only p-doped substrate is relevant).
Diffusions/n-wells always at minimum distance compatible with design rules.
G
S
D
Metal1
n+ source
N+
STI OXIDE
STI oxide
N+
n+ drain
Between n+ diffusions
This FOXFET simulates the situation
where two neighbor NMOS FETs have
terminals at different potential
SUBSTRATE
G
S
D
n+ source
N+
N WELL
n+ drain
STI OXIDE
STI oxide
N+
SUBSTRATE
S
G
D
n+ source
N+
N WELL
n+ drain
IN2P3 school, May 2005
Between nwell and n+ diff
Metal1
Polysilicon
STI OXIDE
STI oxide
N+
N WELL
SUBSTRATE
Federico Faccio - CERN
This FOXFET simulates the situation
where an n-well at Vdd is close to an
NMOS FET with diffusion at gnd
(common case for logic circuits)
Between nwells
This FOXFET simulates the situation
where two n-wells at different potential
are close to each other (this happens
only in analog, and seldom)
86
Leakage between NFETs
1.E-06
ann 2 weeks
65 Mrd
1.E-07
5 Mrd
3 Mrd
pre-rad
Ids (A)
1.E-08
1.E-09
FOXFET with W=200um
Vgs=2.5V during irradiation
1.E-10
1.E-11
1.E-12
1.E-13
0
5
10
15
20
25
30
35
40
Vgs (V)
Leakage increases, but stays in the nA range.
This should not impact digital circuits (if their size is not in the 106 NFETs)
For analog design, issue only for very low current applications
IN2P3 school, May 2005
Federico Faccio - CERN
87
Leakage NFET-Nwell (1)
FOXFET with W=200um
Vgs=2.5V during irradiation
L = twice minimum allowed
L = minimum allowed
1.E-03
1.E-03
1.E-04
1.E-04
1.E-05
1.E-05
Ids (A)
1.E-07
1.E-08
1.E-09
1.E-07
1.E-08
1.E-09
1.E-10
1.E-10
1.E-11
1.E-11
1.E-12
1.E-12
0
10
20
30
40 50 60
Vgs (V)
70
80
90 100
pre-rad
100krad
3Mrad
40Mrad
annealing
1.E-06
Ids (A)
pre-rad
100krad
3Mrad
40Mrad
annealing
1.E-06
0
10
20
30
40
50 60
Vgs (V)
70
80
90 100
The leakage level measured should not compromise the functionality of
digital circuits (if their size is below a few 100.000 gates) even if space is
left at minimum. Leakage peaks at about 3Mrd – as for NMOS leakage!
Analog designers should be careful to this possible source of failure.
IN2P3 school, May 2005
Federico Faccio - CERN
88
Leakage NFET-Nwell (2)
N-well
Without guardring
Vdd
N-well
N-well
Partial guardring
Three layout approaches have been tested in
small logic blocks:
-Standard (no separation)
-Full p+ guardring
-Guardring only where possible (without
cutting poly connections between N and
PFETs)
Irradiation results show that partial guardring
is good enough to improve tolerance
considerably
IN2P3 school, May 2005
Vdd
Full guardring
1.0E-04
1.0E-05
NoGuard
FullGuard
PartialGuard
I (A)
Vdd
1.0E-06
1.0E-07
Federico Faccio - CERN
1.E+04
1.E+05
1.E+06
1.E+07
1.E+08
TID (rad)
89
Leakage Nwell-Nwell
1.E-03
1.E-04
1.E-05
Ids (A)
1.E-06
1.E-07
1.E-08
prerad
100krad
3Mrad
65Mrad
ann 2 weeks
1.E-09
1.E-10
1.E-11
FOXFET with W=200um
Vgs=2.5V during irradiation
1.E-12
0
5
10
15
Vgs (V)
20
25
30
The leakage increase is maximum at about 3Mrd (as for NFET leakage!).
This leakage has to be considered when designing analog circuits –
especially when using N-well resistors or triple-well transistors.
IN2P3 school, May 2005
Federico Faccio - CERN
90
Forward-biased diodes
Devices typically used in bandgap reference voltage circuits.
Two different layout used for the same area of diffusion (STI-bound):
-Small perimeter (one large diffusion)
-Large perimeter (parallel of large number of narrow fingers).
Irradiation performed under direct bias = 0.7V
Large perimeter
1.E-02
1.E-03
1.E-02
1.E-04
1.E-04
1.E-05
1.E-05
1.E-06
65 Mrd
1.E-07
25 Mrd
1.E-06
3 Mrd
1 Mrd
500 krd
1.E-09
500 krd
1.E-10
5 Mrd
1.E-08
1 Mrd
1.E-09
25 Mrd
1.E-07
5 Mrd
1.E-08
65 Mrd
I (A)
I (A)
Small perimeter
1.E-03
pre-rad
pre-rad
1.E-10
1.E-11
1.E-11
1.E-12
0.3
0.4
0.5
0.6
Vforward (V)
0.7
0.8
0.3
0.4
0.5
0.6
Vforward (V)
0.7
0.8
Use the smallest perimeter possible in all cases, and at largest injection level!
IN2P3 school, May 2005
Federico Faccio - CERN
91
Diffusion resistors
9 Two n+ diffusion resistors designed and
measured:
z
z
Minimum width, nominal R=15.03 KΩ
W=1.5um, nominal R = 2.52 KΩ
9 Measured values 17.12 and 2.52 KΩ
(tolerance on minimum width is larger!)
9 Irradiated up to 65Mrd, with changes of
the order of 1%
IN2P3 school, May 2005
Federico Faccio - CERN
92
Why all previous results are worst case (1)
1. Relevance of dose rate
All results shown refer to high “laboratory” dose rates typically above 20krd/min.
In real applications, the dose rates will be 2-3 orders of magnitude below. Since
the radiation response is governed by the combination of oxide trapped charge
and interface states, the overall damage depends on temperature and dose rate.
1.E-03
1.E-06
1.E-04
1.E-05
1.E-06
Ids (A)
Leakage (A)
1.E-07
1.E-08
1.E-07
1.E-08
1.E-09
1.E-09
1.19krd/min
1.E-11
27.5krd/min
1.E-10
0.E+00
prerad
HDR
LDR
1.E-10
1.E-12
0
2.E+06
4.E+06
6.E+06
8.E+06
1.E+07
5
10
15
Vgs (V)
20
25
30
TID (rad)
0.48/0.12 NFET leakage at different dose
rates: the increase is considerably smaller,
and reaches the peak value at lower TIDs,
for the smaller dose rate
IN2P3 school, May 2005
Current of FOXFETs with n-wells (size=200/min)
at two very different dose rates: 24.4krd/min
(HDR) and 65rd/min (LDR). Again, current is
much smaller at low dose rate.
Federico Faccio - CERN
93
Why all previous results are worst case (2)
1. Relevance of dose rate
All results shown refer to high “laboratory” dose rates typically above 20krd/min.
In real applications, the dose rates will be 2-3 orders of magnitude below. Since
the radiation response is governed by the combination of oxide trapped charge
and interface states, the overall damage depends on temperature and dose rate.
TEST circuit: 16kbit SRAM designed with a
memory generator (commercial design, no ELT or
guardring, and dense layout).
1.4
1.2
VDDmin [V]
1
0.8
0.6
LDR
0.4
HDR
HDR+annealing
0.2
0
0
2
4
6
Total Ionising Dose [Mrad]
IN2P3 school, May 2005
8
10
VDDmin is the minimum supply voltage for errorfree functionality at 10MHz. Before irradiation,
VDDmin is about 875mV. With irradiation at
24.4krd/min (HDR), VDDmin increases and peaks
at 1-6Mrd, to decrease when the TID reaches
10Mrd. With irradiation at 65rd/min (LDR), the
increase in VDDmin is very marginal.
Federico Faccio - CERN
94
Why all previous results are worst case (3)
1. Relevance of applied bias
All results shown on NFETs and FOXFETs have been obtained with DC “worst
case” bias:
-NFETs: all terminals grounded except gate at Vdd (1.5V)
-FOXFETs: all terminal grounded except gate at 2.5V
Bias has an important influence on the degradation, as shown in the results below
from M.Silvestri (Dept. of Information Engineering, University of Padova, Italy):
the radiation-induced increase in leakage depends strongly on applied bias.
I/O NFETs
Core NFETs
1.E-05
1.E-06
0.48/0.12 Vgs=1.5V
1.E-08
0.36/0.24, Vgs=0V
0.36/0.24, Vgs Pulsed
1.E-07
I leak (A)
I leak (A)
1.E-06
0.48/0.12 Vgs=0.75V
0.48/0.12 Vgs=0V
1.E-07
0.36/0.24, Vgs=Vdd
0.36/0.24, Vgs=Vdd/2
1.E-08
1.E-09
1.E-10
1.E-11
1.E-09
1.E-12
pre-rad
1.E+05
1.E-10
1.E+05
pre-rad
IN2P3 school, May 2005
1.E+06
1.E+07
1.E+08
1.E+09
TID (rad)
1.E+06
1.E+07
TID (rad)
1.E+08
1.E+09
Pulsed=digital-like (1MHz, 50% duty cycle)
Federico Faccio - CERN
95
Single Event Upset
/bit)
Sensitivity has been tested twice:
-With protons on a sample provided by the foundry (16k x 16 SRAM) in 2001
-With heavy ions on a 16kbit SRAM designed using the commercial memory
generator
1.E-07
-No SEL ever observed
Cross-section (cm
2
8.E-14
s [cm2/bit]
6.E-14
4.E-14
1.E-08
0.13um SRAM @ 1.5V
1.E-09
Weibull 0.13SRAM
0.25um SRAM
Weibull 0.25SRAM
All bits 0
2.E-14
1.E-10
0
All bits 1
5
10
15
20
25
30
35
40
45
50
2
LET (MeV/cm mg)
0.E+00
0
10
20
30
40
Proton Energy [MeV]
50
60
70
Custom SRAM irradiated with Heavy Ions,
and compared to the cross-section measured
Foundry sample irradiated with monofor the SRAM in the “CERN Radiationenergetic protons. The cross-section curve
Tolerant” library in 0.25um. The 130nm
has been obtained with Vdd=1.5V, and
SRAM has a much lower LET threshold. In
shows no significant difference for “all 1” or
an LHC-like environment, this translates in
“all 0” patterns, and large sensitivity at low
increased sensitivity (15-30 times larger!).
energy
96
IN2P3 school, May 2005
Federico Faccio - CERN
Summary for designers (1)
9
Analog design
z
Core FETs ok, however
• Avoid narrow transistors (RINCE)
• Use ELTs for devices in weak/moderate inversion
z
Isolation
• Neighbor NFETs: ok unless very small current
• Nwells surrounded by guardring if the Nwell is sensitive to
leakage (small current) and if neighbor NFETs are sensitive to
small leakage
• Generally enclose with guardrings nodes sensitive to small
leakage currents
z
Diodes
• Use forward-biased diodes with max perimeter/area ratio and
at max injection
z
I/O FETs are quite sensitive to TID, expect large leakage
and Vth shifts
IN2P3 school, May 2005
Federico Faccio - CERN
97
Summary for designers (2)
9
Digital design
z
Core gates
• Use of the commercial library cells should be OK unless
the circuit requires the ultimate performance from the
technology. This still needs to be confirmed by a real
demonstrator.
• If custom cells needed, same recommendation as for
analog design
z
I/O cells
• The suitability of using the commercial library cells still
has to be assessed – only 1 test on few cells has been
made so far (result good, but not very representative)
IN2P3 school, May 2005
Federico Faccio - CERN
98
Summary for designers (3)
9
Single Event Effects
z
SEU
• The sensitivity is larger than for the 0.25um technology,
and mitigation techniques are required for SLHC
applications
z
SEL
• No evidence in any test, and very unlikely due to
technology (dopings, STI) and voltage supply.
Nevertheless, limit resistance to Vdd and Gnd by placing
always lots of contacts across the circuit.
z
SEGR
• Although this should not be an issue, sensitivity will be
tested later in 2007 (with an heavy ion irradiation on
dedicated test structures).
IN2P3 school, May 2005
Federico Faccio - CERN
99
130nm technology node for HEP
Outline
9 The 130nm technology node
9 Radiation effects in samples from
different Foundries
9 CERN Foundry Service in 130nm
z
z
z
Specific features
Radiation effects
Design Kits
IN2P3 school, May 2005
Federico Faccio - CERN
100
Well proximity effect
9
High energy well implants result
in significant scattering of
dopants at the well edge,
increasing the net doping level
and hence threshold voltage for
FETs near a well edge
z
z
The effect is modeled by 1
parameter (which is extracted
for post-layout simulations), for
estimation – simulation will not
be very precise, and it will not
depend on the specific layout
The use of good design
practices for sensitive devices is
necessary – place the sensitive
devices far from the wells
Well
FET1
FET2
Vth1 different than Vth2!
IN2P3 school, May 2005
Federico Faccio - CERN
101
Isolation-induced stress
9
STI induces compressive
strain in silicon, which alters
mobility of FETs
z
z
z
Ids increases for NFETs,
decreases for PFETs
This effect is modeled for
simulation purposes with two
parameters, which can be hand
calculated (geometric size to
measure in the layout)
Extraction from layout is fully
supported for post-layout
simulation
IN2P3 school, May 2005
Federico Faccio - CERN
Effect depends on distance
of channel from STI
STI
STI
Compressive strain
102
Gate damage due to plasma charging (1)
9
9
9
Many processes step for wafer manufacturing take place in a plasma
Isolated metal stripes in a non-uniform plasma get charged at different
potentials. The wafers substrate is at one only potential (typically
grounded), and an electric field builds up in the oxide isolating the
metal from the substrate
In the absence of any current from substrate to metal, large voltage
differences can be reached – exceeding breakdown of the isolation
oxide if this is thin (such as the gate oxide)
Non-uniform plasma
NonSi wafer
IN2P3 school, May 2005
Federico Faccio - CERN
Metal stripes
Isolation oxide
103
Gate damage due to plasma charging (2)
9
9
9
9
Before the voltage differences reaches breakdown, a current starts flowing
across the thin oxide (Fowler-Nordheim tunneling). Current is from electrons
injected from substrate to metal
This current roughly needs to compensate for flux of positive ions on metal from
the plasma. The larger the flux, the larger the current required
If large area of metal is exposed to ions, the flux on the metal is large. If the
metal is connected to a small area of thin oxide, the FN tunneling current per
unit area needs to be large => large voltage across the gate oxide THIS IS
CALLED AN ANTENNA!
This condition leads to damage to the gate oxide, with consequences on the
transistor performance => Vth shift, gate dielectric leakage and increased oxide
reliability failure
Large ion flux from the plasma
Thin oxide area where FN
tunneling current flows
Large metal area
Pre-Metal Dielectric
Poly
STI
STI
Si wafer
IN2P3 school, May 2005
Federico Faccio - CERN
104
Gate damage due to plasma charging (3)
9
9
To avoid damage, Design Rules limit the allowable ratio of PC/Metal to
thin oxide area. This applies individually to all metal layers (it is not
cumulative across layers)
Solutions to avoid damage:
z
z
Add “tie-down” diodes, connected via M1 to the PC (in this case, the limit
ratio from the design rules still applies but the tie-down diode area sums –
with multiplication coefficient – to the thin oxide area). At wafer processing
temperatures, diodes are very conductive and allow current to flow from
metal to substrate
Introduce “hops” to next metal level. Before the “hop”, only the area of the
metal already connected to the thin oxide counts for the antenna. When
processing the next metal level, only the (small) area of the “hop” counts
for antenna.
TieTie-down
diode
Long metal1 line
IN2P3 school, May 2005
Metal 1
connection
Metal 2
“hop”
Long metal1 line
Federico Faccio - CERN
Metal 1
connection
105
Copper dendride formation
9
This is a complex effect that can
take place during processing of
copper metal layers
z
z
9
Potential differences built across
wells (n and p- wells) can
generate currents in the solutions
where processing related to a
copper metal layer is taking place
The induced currents will form
“dendrides” around the copper
metal line connected to the well,
which can lead to shorts to
neighbor lines in the same metal
level
To prevent this to happen, Design
Rules have been introduced
z
Large well
Small metal
connected
to the well
Dendrides
Ratio of metal to well area has to
be large
IN2P3 school, May 2005
Federico Faccio - CERN
106
Pattern Density
9
STRICT requirements exist for density of each of the
following layers: RX (active area), PC (poly), all metals
Requirements are both for the full chip (global rules)
and for any small area of it (local rules)
z
z
9
9
Global rules set the limit – upper and lower density – for
the full chip
Local rules set the limit – upper and lower density – for
areas about 100um wide stepped by about 50um across
the full design
After tape-out, automatic routine at the Foundry will fill
in all layers (“filling”), and produce holes in copper
layers (“cheesing”). This is unavoidable
Some consequences:
z
z
Example:
It is not possible to place “exclude” shapes over area
where filling is not wanted. To prevent random metal
shapes to be placed over sensitive portion of the design
(for instance, where matching is important), cover the
sensitive area with uniform metal
Some designs that produce excessive local density CAN
NOT be manufactured. Example: regions with too large
usage of RX. This error is spotted by running a check that
predictively estimates the final densities after filling.
IN2P3 school, May 2005
Federico Faccio - CERN
Large array of large
transistors (for instance,
a large current mirror)
100um
9
150um
Local RX density migh be
eccessive here!
The design HAS to be
modified or it will be
rejected from fabrication!
107
Check procedure before tape-out
9
Due to antenna, dendride formation, filling and other complex
rules, the checking procedure before tape-out requires several
steps just for Design Rule Check (DRC)
z
z
z
z
Check “ground rules” (GR), or rules mainly related to geometrical
requirements
Check floating gates and antennas to ensure rules related to
plasma damage and dendride formation are satisfied
Check global density
Check (predictively) local density – the design can be filled at the
Foundry to meet all local density requirements
All these checks can be performed with Assura (Diva has very
limited use for checking). Calibre should also be able to do the
job, but we have no experience with it
9 NB: procedure to get “waivers” as we were used with the
quarter micron is now much heavier and given the complexity of
the technology the risk is also higher
9
IN2P3 school, May 2005
Federico Faccio - CERN
108
Digital design flow
The complexity of the technology, its high performance, and the
need for minimizing expensive prototyping cycles drives the
need for sophisticated tools for digital design
9 A full library is available from a commercial supplier, but how to
merge it with the foundry design kit in an optimized and
standard design flow?
9 The task of rationalizing the digital design flow and to providing
a common platform for design tools has been outsourced to
Manhattan Routing Inc. (MRE)
9
Chip design
(Verilog)
Physical
implementation
Physical
verification
Manhattan Routing Digital Design Kit
IN2P3 school, May 2005
Federico Faccio - CERN
109
The MRE Digital Design Flow in detail
Synthesis
dc_shell
Placement
psyn_shell
Post Route Opt
pks_shell
ECO
First Encounter
Floorplanning
First Encounter
Placement Opt
pks_shell
Hold Fx
pks_shell
IPO Route
nanoroute
Clock Tree Gen
fects
IPO Route
nanoroute
Extraction
Fire & Ice (qx)
Post CTS Opt
pks_shell
Extraction
Fire & Ice (qx)
SI Analysis
celtic_ndc
Route
nanoroute
SI Analysis
celtic_ndc
Timing Analysis
pt_shell
Extraction
Fire & Ice (qx)
Timing Analysis
pt_shell
SI Analysis
celtic_ndc
Timing Analysis
pt_shell
Manhattan Routing Inc.
IN2P3 school, May 2005
Federico Faccio - CERN
GDS Editor
dfII
Physical Verification
calibre
110
User point of view
The design flow uses specific tools
9 Users shall use the same tools, and the appropriate versions of
those, to be able to run the flow correctly
9
z
9
Tools and versions used in the latest release of the Design Kit
(V1.3):
1-week courses with limited attendance (10 people) are
organized periodically at CERN. 3 such courses have already
been given by MRE personnel, and another one is planned
IN2P3 school, May 2005
Federico Faccio - CERN
111
130nm technology node for HEP
Summary
9
9
9
CERN will offer a Foundry Service, as already done in the 250nm
node, for the 130nm
both CMOS (mixed-signal) and BiCMOS (with SiGe HBTs) can be
offered
Radiation tolerance of the CMOS technology:
z
z
z
9
9
9
9
It has been tested extensively, and a dedicated HBD digital library does not
seem needed
A set of recommendation to the use of the technology for designs that have
to operate in a radiation environment is available
The stability of the natural radiation tolerance will be monitored frequently
Cadence-based PDK is available
Digital library is available
Both PDK and library are distributed within a “Design Kit” from
Manhattan Routing Inc., together with a Design Flow for complex
digital design
CERN-organized MPW runs will start as soon as demand for silicon is
sufficient to bring price to competitive level. In the meanwhile, frequent
runs are accessible via a broker (CERN can act as contact for such
runs)
IN2P3 school, May 2005
Federico Faccio - CERN
112