Transcript Document

The SVX4 Readout Chip
Vertex 2002
M. Garcia-Sciveres, Lawrence Berkeley National Lab.
SVX’
1990
SVX2
1996
SVX3
1998
SVX4
2002
The SVX Family of chips
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
1
Credits & History
• Design Team:
– LBNL: Brad Krieger (lead designer), Jean-Pierre Walder (ADC), Henrik von
der Lippe (I/O pads), Emanuelle Mandelli (full chip simulation)
– FNAL: Tom Zimmerman (preamp & pipeline), Jim Hoff (pipeline logic)
– U. of Padova: Stefania Alfonsi (FIFO)
• Timeline
– July 2000: First SVX4 proposal meeting following studies at LBNL
– May 2001: Common CDF/D0 specifications supplied to designers.
Preliminary design review.
– October 2001: Final Design Review
– April 2002: Design submitted for fabrication
– June 10, 2002: Wafers delivered to FNAL
– June 12, 2002: First chip tested at LBNL, basic functionality verified.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
2
Why SVX4?
• Present CDF and D0 detectors not expected to survive
beyond 4-6 fb-1
• Noise increase in SVX3 and SVX2 chips is important
limitation on lifetime
• Not possible to re-order existing chips to build new
detectors
• Take advantage of new Quarter Micron technology and
work done for LHC experiments.
• Functionally, SVX4 ~
= SVX3 + additional interface logic
to make it optionally heave like SVX2 chip.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
3
Review of SVX3
(presented at Vertex 1998)
ADC
8
Charge sensitive
amplifier with
adjustable
risetime
Dual-ported
analog
pipeline for
dead-timeless
operation
Wilkinson
Zero-suppressed
type ADC
128 channels readout on 8-bit
parallel bus
in parallel
• Low resistivity substrate exploited for ground distribution
and digital/analog isolation
• Real time pedestal subtraction built into ADC
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
4
SXV4 features
• Use same floor plan as SVX3
• Most analog circuits redesigned at schematic level- full
custom layout
• Digital circuits made using library parts from rad-hard
library developed for ATLAS pixel chip, based on RAL
quarter micron library.
• Use low resistivity substrate as in SVX3
• Add on-chip decoupling capacitance
• Design relies heavily on fantastic accuracy of high volume
commercial process simulation tools – this is THE biggest
change relative to traditional rad-hard electronics.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
5
Results
• Full SVX4 chip was simulated before submission. The
problem is no longer “can you simulate it” but rather “what
to simulate”.
• First full size chip fabricated is essentially a usable chip:
The number of changes needed for the next SVX4 submission is less
than the number of changed made for the SVX3 production run.
• The things that need to be fixed were not caught by the
simulation because the right things were not simulated.
• Every item could be simulated after the fact, including a
very subtle ADC effect (shown later).
• Radiation tolerance to spare. SEU x-section lower than
SVX3- same register cells as ATLAS but 100 times fewer
of them.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
6
Analog performance
ENC (electrons)
svx3D
4MRad
SVX4
10MRad
3500
3000
2500
2000
1500
1000
500
0
0
10
20
30
40
50
Input Capacitance (pF)
100ns integration time, 70ns 0-90% rise time,
290uA/channel preamp current
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
7
Zero-Suppressed Data Out “FIFO”
• SVX3 “FIFO” is full custom, dynamic
logic circuit with asynchronous
collapse to achieve zero suppression.
• This was the only way to do it fast
enough in 0.8mm.
• 0.25mm is SO MUCH faster that we
could instead use a fully synchronous,
automatic place and routed VHDL
circuit with no collapse phase- zero
suppression happens in real time
during readout.
• At design readout speed if 53MHz it
works exactly except for some special
low occupancy hit patterns in which
case a single extra readout clock is
used. This OK for us.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
8
Full chip Simulation
• >300K transistors
• TIME-MILL
program with some
tricks
• Save DC points of
programmable
registers
• About 1hr per ms
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
9
ADC Comparator channel-to-channel scatter
ADC counts (pedestal)
Measurement: Each line is a different comparator bias current
Less bias
More bias
Channel number
• Pedestal shifts but gain is the same for all traces
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
10
ADC scatter cause
• All comparators identical at schematic and layout levels.
• Quarter micron circuits it turns out are particularly sensitive to tiny
threshold mismatch in “matched” (at layout level) transistors. – This is
NOT the same thing as lot-to-lot global variations or “corners”
• TSMC in fact provides data on this type of mismatch for every
transistor parameter (for IBM data comes from a thesis)
• With these data one can run a Monte Carlo circuit simulation were the
parameters are wiggled for selected transistors.
• We wiggled other parameters, not just Vth, but found that Vth is the
only one that matters.
• Interestingly the ATLAS FE chip was also bitten by this bug, which led
to a higher than desired threshold dispersion in current prototype. We
copied the MC simulation idea from them.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
11
MC Simulation of Channel-to-Channel Variations
ADC counts
Ramp-up Pedestal Variation: Measured vs. Monte-Carlo
Simulation of Vth Using TSMC Process Data
70.0
65.0
60.0
55.0
50.0
45.0
40.0
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
58 58
23 25
20
17
13 13
6 5.37.2
4uA
10uA
Measured SPREAD
Simulated, 4 transistor MC
Nov. 7, 2002
68
2.22.63.8
16uA
Bias current
d16_10
d16_4
Measured DELAY
Simulated, 4 transistor MC
Simulated, all transistor MC
Simulated, all transistor MC
The SVX4 Chip -- M. Garcia-Sciveres
12
Conclusion
• SVX3 circuit was converted to .25mm with some
enhancements in a little over 1 year.
• SVX4 significantly outperforms SVX3.
• Full chip prototype in hand 2 years after first we though of
doing it is already closer to a production chip than the final
SVX3 prototype was.
• Really a new (to us) way to design ICs. Use state of the art
industry tools for high volume process. Full chip
simulation leading to very high confidence submission.
Nov. 7, 2002
The SVX4 Chip -- M. Garcia-Sciveres
13