1 / 28

A Proposal for the Readout of the NA62 LKr Calorimeter

A Proposal for the Readout of the NA62 LKr Calorimeter. New NA62 LKr readout. using EXISTING HARDWARE as much as possible. Preamplifiers (kept cool). Transceivers and Cables (faulty devices should be repaired). CPD to be upgraded to 1 MHz readout rate (3*10 6 events/burst?).

verdad
Download Presentation

A Proposal for the Readout of the NA62 LKr Calorimeter

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Proposal for the Readout of the NA62 LKr Calorimeter B.Hallgren, M.Piccini and H.Wendler

  2. New NA62 LKr readout using EXISTING HARDWARE as much as possible • Preamplifiers (kept cool). • Transceivers and Cables (faulty devices should be repaired). • CPD to be upgraded to 1 MHz readout rate (3*106 events/burst?). • The new CPD (nCPD) must be able to operate together with the present CPD. • This means that a staged replacement procedure can be implemented module by module. • The CPD module mechanics is kept. • Fastbus crates and cooling are kept. • Optical links (high failure rate no spare parts) are being replaced by the Gigabit Smart Link Module (SLM). • Data Concentrator to be replaced with Network Switch and FPGA or PC FARM. B.Hallgren, M.Piccini and H.Wendler

  3. Calorimeter Pipeline Digitizer CPD • 32xCPDAS cards -> kept • Analog Trigger Subcard Digital Subcard -> new design. • CPD Motherboard -> remake digital parts but carefully copy the analog layout. B.Hallgren, M.Piccini and H.Wendler

  4. Present Readout of the CPDAS • The circuit which connects the ADC to the pipeline memory the MIGAgatearray operates together with the address and control signals in the following way: • The memory operation cycle is divided into 2 x 25 ns time periods performed in parallel on all channels. • The 1st period is used for 2 ADC words that are written into the memory every 50 ns (Circular Buffer) via the MIGA. • The 2nd period can be used to write or read two words from the memory (Linear Buffer) via the MIGA. • The readout from the MIGA done is done channel by channel and takes 64 x 0.8 = 51.2 µs + overhead. B.Hallgren, M.Piccini and H.Wendler

  5. new CPD BLOCK DIAGRAM B.Hallgren, M.Piccini and H.Wendler

  6. New Readout of the CPDAS • In the new CPD each MIGA is connected to a FPGA (per 4 CPDAS or 8 channels). The readout of 2 words is done all channels in parallel and takes 100 ns. A complete event consisting of 8 words takes 400 ns. • Transfer of data from the channel FPGA to the module FPGA is done with 8 x LVDS 12 bit buses operating at 80 MHz (using ADC clock). • An event consisting of 8 x 12 bit words/channel will be transferred in 800 ns (8 channels) from the channel FPGA (cFPGA) to the module FPGA (mFPGA). • Transmission to the SLM is done with LVDS board 12bits @ 80 MHz. • Update of FPGAs and loading of CPDAS constants can be done via the SLM using the Gigabit Ethernet (bidirectional) and two lines of the LVDS bus. B.Hallgren, M.Piccini and H.Wendler

  7. Module with 2 SLM mounted on the back of the LKr Rack 2xSLM CPD-LVDS card replacing optical links B.Hallgren, M.Piccini and H.Wendler

  8. Mass Market Key Components LAPTOP DDR2 SODIMM 10/100/1000Ethernet FPGA B.Hallgren, M.Piccini and H.Wendler

  9. Field Programmable Gate Array FPGA • No Gates anymore but Logic Elements. • Hardware blocks such as Memories, Transceivers, Programmable Delay Lines, LVDS receivers, ADC, JTAG, Logic Analyzers … • Software blocks (Encrypted) Ethernet and DDR2 memory controllers .. • Design is hardware oriented using “MegaWizzard Plug-In Manager”. • FPGA system clock is only 100 to 200 MHz due to routing delays. • Latest FPGAs are available in 65nm technology (faster, smaller packages, lower dissipation and cost). • Example for nCPD in Cyclone III EP3C5_164MBGA 8x8 mm $17 x 8 • EP3C40_780FBGA 29 x 29 mm $133 x 2 (4) • But high-end FPGAs Stratix III such as EP3SL150_780FBGA costs > $ 2185 • The Stratix III EP3SL340_1517FPGA costs > $11168 B.Hallgren, M.Piccini and H.Wendler

  10. Double Data Rate 2 DDR2 Memory • Analog technology using capacitors as storage elements • With no power the memory content is be kept for a few sec @25 ºC • DDR2 memories was introduced year 2003 • 2 Gbytes of DDR2 @ 667MHz SODIMM costsCHF 47(14/4-08) • Burst speed is 667 MHz but only for 4 or 8 clock periods! • Can operate with R/W cycle times of about 100 ns. • SODIMM DDR2 memory works with 64 bits, but for example the ALTERA Cyclone III DDR2 controller uses words of 256 bits. • Data from one CPD channel is stored in 128 bits to simplify addressing. • 64 CPD channels requires 1 kB DDR2 memory • 2 Gbytes/CPD = 2 sec of NA62 data taking @ 1MHz trigger rate Kingston ValueRAM DDR2 667MHz CL5 - 2 GBKVR667D2N5/2G - 2048 MB - DDR2 - PC2-5300 - DIMM - Bus 667 MHz B.Hallgren, M.Piccini and H.Wendler

  11. Gigabit Ethernet • 1 Gigabit Ethernet with 4 twisted pairs was standardized 1999. • Fully duplex with each port using 17 analog levels at 125 MHz. • Present price of Gigabit Ethernet (10/100/1000BASE-T) • Components per module port • Marvell 88E1111 + RJ45 jack with trafo CHF 16.- • CAT 6 cable 20 m Halogen free CHF 47.- • Price/port in a Gigabit managed Network Switch CHF 100.- • 10 Gigabit Ethernet using 4 twisted pairs was standardized 2006 Price of 100/1000/10000 BASE-T is what 2011? B.Hallgren, M.Piccini and H.Wendler

  12. Trigger Timestamp to the nCPD • In contrast to LHC there is no TØ in the NA62 • The same type of trigger as the old system is needed for the nCPD • A timestamp which gives the time or address of the pipeline DDR2 memory to read out • Distribution of the timestamp can be done using the FASTBUS backplane as a Broadcast Message • The connections between crates are OK with Segment Interconnect • The VME TIC connections must be changed for 1 MHz B.Hallgren, M.Piccini and H.Wendler

  13. Trigger Outputs Trigger outputs 10 m cable Trigger outputs 10 m cable B.Hallgren, M.Piccini and H.Wendler

  14. Trigger Outputs • Fast Trigger summing outputs are available at 10 m from the LKr crates, see Vienna Trigger Descriptions Differential analog currents 2 x 20mA 4 in X and 4 in Y per CPD Special backplane in the crate for the analog trigger partial summing In total 864 in X and 864 in Y outputs B.Hallgren, M.Piccini and H.Wendler

  15. Example of using the nCPD in NA62 (based on the ANTARES experiment) B.Hallgren, M.Piccini and H.Wendler

  16. NA62 nCPD SLM READOUT • Each nCPD has one LVDS link to SLM with 120 Mbytes/s. • The SLM can receive up to 480 Mbytes/s (4 links from nCPD). • Each SLM has one bidirectional Gigabit Link = 80 Mbytes/s (max). • Data reduction needed (150 active channels of 13200 in average): • Example of readout of the nCPD: • Data reduction (like the Data Concentrator but with Network Switch and TELL1 clone). • “JPEG” picture of the calorimeter. B.Hallgren, M.Piccini and H.Wendler

  17. Data reduction like the Data Concentrator • For each event the nCPD determine the channels which have values above threshold and store these channels in a bit pattern consisting of 8 bytes per nCPD and 32 bytes per SLM. • SLM assembles one GbE packet with i.e. 25 events. • Use Commercial Network Switch, i.e. as HP ProCurve: • to distribute packets of the same event from all CPDs to one HALO FPGA processing node. • to send data from clusters to DAQ channels. • Halo processing is still difficult, two alternatives: • With 12 TELL1 clones with 4 special subcards the HALO processing could be made in FPGAs by loading a matrix consisting of 128 x 128 logic elements, (better than custom hardware but needs electronics engineers). • With a PC farm and Gigabit Switch (should be more flexible and easier to program). • The channels to read out (the results from the HALO processing) are sent to all nCPDs as coordinates and readout. B.Hallgren, M.Piccini and H.Wendler

  18. Data reduction (based on ANTARES) B.Hallgren, M.Piccini and H.Wendler

  19. “Example” of HALO PROCESSING Parameters with SLMs and Network Switch B.Hallgren, M.Piccini and H.Wendler

  20. Other use of the FPGAs in the nCPD B.Hallgren, M.Piccini and H.Wendler

  21. “JPEG” picture readout of LKr as independent check of data quality The FPGA calculates the mean value xm of 8 samples xi and the sum Σ(xi –xm)2 White Grey Blue Red 4 colors = 2 bits/channel sent in Jumbo packets of 128 events gives ~ 64 Mbytes/s per SLM. One event of the whole LKr is 3.3 Kbytes + the ~150 channels of the cluster. B.Hallgren, M.Piccini and H.Wendler

  22. Increased resolution with a digital filter? The frequency spectrum of the noise of the LKr electronics is such that it could be efficient to use a digital filter to decrease the bandwidth and thereby increase the resolution for pedestals. FPGAs are very suitable for digital filters. B.Hallgren, M.Piccini and H.Wendler

  23. Examplesof implementation with commercial electronics B.Hallgren, M.Piccini and H.Wendler

  24. Modular Managed Network Switch Modular chassis with dual power supplies with place for up to 12 modules (i.e. 288 ports) The cost is about $14000 for a switch with 144 10/100/1000 ports Example of modules B.Hallgren, M.Piccini and H.Wendler

  25. >$100000 ? B.Hallgren, M.Piccini and H.Wendler

  26. Part of a PC farm with 44 x 1U PC 2.5m 1 PC =$1000 B.Hallgren, M.Piccini and H.Wendler

  27. Discussion • By using large memories and FPGA technology in the nCPD a smart readout can be implemented. • Two cards in the CPD must be replaced to cope with the 1 MHz rate. • This should be straightforward if the old design documentation, schematics and design files are still available. We estimate that about 2 man years are needed? One person mainly for the FPGA design the other for the design and production of the new PCBs. • A staged replacement procedure could be implemented: • The calibration constants for the CPDAS requires that each subcard is preferably put back into its old position on the new motherboard - module by module. • The full capability of the system is only needed when operating at 1 MHz speed. Therefore the network switch and the PC farm should be equipped partially as needed and as late as possible to get the optimum technology i.e. multi-core processors and 10 GbE. • A technical note is still in preparation maybe withTWiki instead. B.Hallgren, M.Piccini and H.Wendler

  28. B.Hallgren, M.Piccini and H.Wendler

More Related