1 / 16

L1Calo upgrade discussion

L1Calo upgrade discussion. Samuel Silverstein Stockholm University. Overview Issues Latency Rates Schedule Proposed upgrade strategy R&D. Upgrade phases (for this talk). Phase I (before calorimeter upgrades): Two projects FPGA-based MCM replacement for PPr

latona
Download Presentation

L1Calo upgrade discussion

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. L1Calo upgrade discussion Samuel Silverstein Stockholm University • Overview • Issues • Latency • Rates • Schedule • Proposed upgrade strategy • R&D

  2. Upgrade phases (for this talk) • Phase I (before calorimeter upgrades): • Two projects • FPGA-based MCM replacement for PPr • CMM++ in CP and JEP, options for • Limited topo algorithms in CMM++ modules • Topological processing in separate TP • Phase II (digital calorimeter readout to USA15) • "Level 0": topological algorithms on finer-granularity "mini towers" built in RODs, plus muon objects • Fixed, low latency (< 3.2 s) • "Level 1": full calorimeter resolution, plus objects from muon and ID triggers • Longer latency, possibly asynchronous

  3. CP 1 CP 0 CP 3 CP 2 Jet 1 Jet 0 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 SNAP12 Phase I with CMM++ (one example) 1 of 2 160 Mbit/s 160 Mbit/s 12 Run faster? 6 12 160 Mbit/s 6 To CTP Can use half of CP ROIs Quadrant border data To CTP

  4. Topo Processor E/gt/had clusters (CP) 0.2 x 0.2 Jet / SET (JEP) High-speed optical links Muons To CTP Clusters 0.1 x 0.1 Jets Energies

  5. Phase II concept • Latency and rates • Fast L0 (40 MHz in, up to 500 kHz out, <3.2s) • Slower L1 (500 kHz in, <75-100 kHz out, >9.6s) • Level 0: • Finer-granularity "mini towers" with e.g. Et, fine-location of shower maxima, longitudinal profiles, quality flags, etc. • Multi-crate L0Calo processor: sliding-window algorithms to identify features • L0Topo: trigger algorithms using calo and muon objects • L0 ROIs sent to L1Track to gather track candidates • Level 1: • Refined L0 algorithms with full calorimeter data • L1Topo: trigger algorithms using calo, muon, track objects

  6. Phase II concept (Murrough)

  7. Latency issues • Phase I: • Latency envelope 2.5 s • May currently be as few as 10 BCs from limits • Phase II: • L0 trigger limited to 3.2 s (by muons) • Tile and (esp) LAr digital processing add additional latency • At least three sequential multi-Gbit links in L0 RTDP • Need to evaluate and limit all potential sources of latency

  8. Latency issues • Sources of additional latency: • Link choice (SerDes latency) • GBT: 8-10 BC if optimised (S. Muschter), 16 BC (GBT group) • 8b/10b: 5 BC (3.2 Gbit),3 BC (6.4 Gbit) • Upgraded firmware • CPM and JEM merger encoding (and decoding): ~0.5 - 1 BC • CTP upgrade to 80 MHz: 2-3 BC • Topological algorithms themselves • Possible latency savings • Faster FPGAs in upgraded hardware • CMM++ (Virtex E  Virtex 6) • PPr MCM upgrade from ASIC to Spartan 6

  9. Rate issues • Conflicting demands at sLHC • L1 tracker concept relies on high L0 trigger rate • Richard Brenner: >500 kHz • Muon detector cannot handle high L0 trigger rate • < 200 kHz • No obvious options • Upgrade muon electronics • Many inaccessible boards. Years to disassemble and replace • Muon system ignores L0A, reads out after L1A • L1A latency too long. Data lost • "Private trigger" for muons? • Won't work....how do we decide which L0A will be accepted by L1? Data lost. • How do we proceed?

  10. Schedule issues • Phase I: • CMM++ should be ready for deployment by 2015 • Need to start Phase I soon • May be needed before calorimeter upgrade (2017-19) • but discarded when calorimeters are upgraded • Phase II: • Large, completely new system! • Concurrent with calorimeter upgrade (2017-19) • Need to start Phase II soon • Problem: Phase I and II development in parallel • conflicting demands on manpower, organisation, etc

  11. Proposed strategy • Focus on Phase II • "Staged" deployment: • L0Calo/L0Topo in time for calo upgrades • 2017-2019 timescale • Fixed latency, maximum 3.2 s • Run at <75 kHz until L1 is ready • L0Topo compatible with CMM++ outputs • Usable as TP in Phase-1 system (see next slide) • L1 trigger ready for ID upgrade • Increase L0 rate to 500 kHz • Asynchronous processing?

  12. Phase-II "staging" Install with calorimeter upgrades (2017-2019) Install with ID upgrade (2020-)

  13. Proposed strategy • Phase-I development in parallel • Lower initial ambition level • Build and deploy CMM++ as soon as possible • Limited topo algorithms in CMM++ • Muon ROIs not included? • L0Topo compatible with CMM++ outputs? • Use as TP if Phase I delayed, or too much pileup • Early, parasitic testing/integration of L0Topo before rest of L0Calo is ready? • Possible to include muon ROIs?

  14. R&D results • MCM replacement design is progressing well • Spartan 6 FPGA could be used to improve pileup performance, maybe lower latency. • Encouraging work on CPM merger • Readout scheme makes better use of bandwidth, small cost in latency and design size • Use same format for Jets? Maybe add jet ROI energies to threshold bits • New ideas for CMM++ • Virtex 6 HXT adds more logic, plus up to 72 multi-Gbit transceivers (from half that on LXT device) • More optical links open new possibilities for merger topology

  15. R&D results • New results from GOLD • Encouraging link latency results • Experience with new HXT FPGAs • Understanding of phase II HW issues, including power, board density, etc • Stockholm link test board • High-speed links can be run with TTC • Latency results compatible with GOLD 8b/10b measurements • IPG during empty LHC BCs • Similar scheme needed for any 8b/10b link. • New algorithm studies • Many algorithms possible with CMM++ only • Interesting idea to combine jet and energy  transverse mass • Need to focus on questions of bandwidth (vs. processing power): how much information is needed, and where?

  16. Discussion?

More Related