190 likes | 202 Views
This overview discusses the crucial aspects of links and clock distribution aiming at the Technical Design Report (TDR). Explore the trends, challenges, and solutions in implementing high-speed serial links, particularly using FPGA. Learn about FPGA testing, fault analysis, and upcoming developments in clock distribution technology.
E N D
Links & Clock distribution: toward the TDR A. Aloisio, R. Giordano aloisio@na.infn.it, rgiordano@na.infn.it University of Naples ‘Federico II’ and INFN ETD Meeting - CERN - Nov. 17th, 2011
Overview • Toward the TDR • Does a Golden Solution exist ? • What if ... • Making the system working • Manpower • Discussion ETD Meeting - CERN - Nov. 17th, 2011 2
Toward the TDR • Setting the scene of links and clock distributionin the ETD • Many key design elements are still missing (rad dose, partitioning, technology) • Devices non yet available today will be obsolete in a few years from now • Trends, way-of-designing, paradigms survive better, can contribute indeed to the TDR ETD Meeting - CERN - Nov. 17th, 2011
Links & Clocks: taking the golden section • In our view, high speed serial links should be the golden solution to distribute ALSO the clock • In our view, their implementation should rely on FPGA as much as possible IDEA DESIGN ETD Meeting - CERN - Nov. 17th, 2011
FPGA pros and cons • Pros: • Uncountable... • Cons • just one: Rad-tolerance • Rationale • A) test on beam the newest FPGAs • B) keep an eye on the silicon vendors • C) Check what ‘competitors’ do • D) Have ready a benckmark design • E) Keep simulating the background ETD Meeting - CERN - Nov. 17th, 2011
Testing FPGA ‘a-la-carte’ • First test on Xilinx V5 – Jul. 10 • Tested TMR vs. noTMR moderation • Config. Memory cross-section • System cross-section of a real design • Power issue: negligible • TID effects: negligible • Latch-up: None • SEFI: minor effects • VERY detailed fault analysis presently being conducted (see next slide) • Three more testbeams coming: • Dec. 10 (V5-V6) then Mar 2012 (V6), Jul 2012 (V6-V7) • No more beam will be available after that ETD Meeting - CERN - Nov. 17th, 2011
CLB 1 CLB 2 CLB 3 CLB 4 CLB 5 CLB 6 CLB 7 CLB 8 CLB 9 Fault Analysis (from last TB) OPEN PIP: net "GLOBAL_LOGIC0_451" gnd, outpin "XDL_DUMMY_INT_X2Y49_TIEOFF_X2Y49" HARD0 , inpin "data_out(13)_dup_117" DX , inpin "data_out(13)_dup_117" SR , pip CLBLL_X2Y49 SITE_BYP_B6 -> M_DX , pip CLBLL_X2Y49 SITE_CTRL_B3 -> M_SR , pip INT_X2Y49 BYP6 -> BYP_B6 , <<<<<<< pip INT_X2Y49 CTRL3 -> CTRL_B3 , pip INT_X2Y49 FAN5 -> FAN_BOUNCE5 , pip INT_X2Y49 FAN6 -> FAN_BOUNCE6 , pip INT_X2Y49 FAN_BOUNCE5 -> BYP6 , pip INT_X2Y49 FAN_BOUNCE6 -> CTRL3 , pip INT_X2Y49 GND_WIRE -> FAN5 , pip INT_X2Y49 GND_WIRE -> FAN6 , ; • We know exactly where we failed • Routing seems to be the problem most of the time: we can moderate such issues • Analysis is performed by Prof. Violante & Prof. Sterpone (Politecnico Torino, Italy) ETD Meeting - CERN - Nov. 17th, 2011
From the vendor: V6 vs.V5 ETD Meeting - CERN - Nov. 17th, 2011
From the vendor: V7 • Starting with V7, it should/could be possibile to correct-on-the-fly a single SEU. Waiting for more info from Xilinx about that... • Trend is clear: SEU will be handled in the future more and more carefully by specific on-chip logic ETD Meeting - CERN - Nov. 17th, 2011
What BELLEII is doing ? • Help yourself ... ETD Meeting - CERN - Nov. 17th, 2011
Benchmark design • We have it • Fixed latency • Low phase noise on recovered clock • Tested, presented many times, published ETD Meeting - CERN - Nov. 17th, 2011
Background simulation • Vital • Protons? Neutrons? E spectra? • More info needed to drive the tests ETD Meeting - CERN - Nov. 17th, 2011
What if ... • ... FPGAs do not fulfill our expectations on-detector • Keep sending both clock and data • Use COTS SerDes (like National, GOL for read-out only, ...) • Design a Rad-Hard SerDes (see next slide) • If (really) unfeasible, then: • Send clock via specific fibers (see later) • Design low speed SerDes (500Mbit/s) with ACTEL-like devices ETD Meeting - CERN - Nov. 17th, 2011
Designing our own SerDes • We are in touch with IMEC (B) to explore the feasibility of a custom made SerDes tailored to our needs • Nov.22 we will talk to R&D Managers from IMEC for details • About costs: 500k€, about time: 1 year • Too expensive? Not really... EU programs, ESA interested, ETD Meeting - CERN - Nov. 17th, 2011
Sending the clock • We do not like that... • ... but we know how to do it and we have already did a few tests (just in case ...) Jitt = 25 ps RMS Jitt = 9 ps RMS Clock driven by off–dector FPGA, to be send across fiber - 40MHz CERN TTC – 40MHz ETD Meeting - CERN - Nov. 17th, 2011
Making the system working • TDR should define also important guidelines: • In our opinion, ALL links (FCTS, Read-out) should be easy upgradable, BOTH ends (TX,RX) • Then, they should be always hosted on mezzanine cards, BOTH ends (TX,RX) ETD Meeting - CERN - Nov. 17th, 2011
Links should talk each other • Like the ride from Eden to Babel ? • In our view, one team does one link front-to-back, see ‘Facts and Opinions ...’ presentation at Elba • Crucial issue, to be discussed in the view of TDR Tower of Babel - Bruegel The Garden of Eden - Bosch ETD Meeting - CERN - Nov. 17th, 2011
Manpower ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ETD Meeting - CERN - Nov. 17th, 2011
Discussion …. • …. Is now open …. ETD Meeting - CERN - Nov. 17th, 2011