1、Nov. 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,1,The SVX4 Readout Chip Vertex 2002 M. Garcia-Sciveres, Lawrence Berkeley National Lab.,SVX2 1996,SVX3 1998,SVX4 2002,SVX 1990,The SVX Family of chips,Nov. 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,2,Credits & History,Design Team: LBNL: Brad Krieger (
2、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
3、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 Ch
4、ip - M. Garcia-Sciveres,3,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 d
5、one 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,4,Review of SVX3 (presented at Vertex 1998),Low resistivity substrate exploited for ground distribution and digital/analog isol
6、ation Real time pedestal subtraction built into ADC,ADC,8,Charge sensitive amplifier with adjustable risetime,Dual-ported analog pipeline for dead-timeless operation,Wilkinson type ADC 128 channels in parallel,Zero-suppressed readout on 8-bit parallel bus,Nov. 7, 2002,The SVX4 Chip - M. Garcia-Scive
7、res,5,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-
8、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,6,Results,Full SVX4 chip was simulated before submission
9、. 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 fixe
10、d 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.
11、 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,7,Analog performance,100ns integration time, 70ns 0-90% rise time, 290uA/channel preamp current,Nov. 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,8,Zero-Suppressed Data Out “FIFO”,SVX3 “FIFO” is full custom, dynamic logic circuit with asynchronous collapse t
12、o 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 53
13、MHz 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,9,Full chip Simulation,300K transistors TIME-MILL program with some tricks Save DC points of programmable regist
14、ers About 1hr per ms,Nov. 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,10,ADC Comparator channel-to-channel scatter,Channel number,ADC counts (pedestal),Measurement: Each line is a different comparator bias current,Less bias,More bias,Pedestal shifts but gain is the same for all traces,Nov. 7, 2002,Th
15、e SVX4 Chip - M. Garcia-Sciveres,11,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 varia
16、tions 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, b
17、ut 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,12,MC Simulation of C
18、hannel-to-Channel Variations,Measured SPREAD,Simulated, 4 transistor MC,Simulated, all transistor MC,Measured DELAY,Simulated, 4 transistor MC,Simulated, all transistor MC,Bias current,ADC counts,Nov. 7, 2002,The SVX4 Chip - M. Garcia-Sciveres,13,Conclusion,SVX3 circuit was converted to .25mm with s
19、ome 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.,
