CMS and the Number Sorter Presented to you by: Bobby Scurlock Faculty Mentor: Darin Acosta

Event = total collection of detector measurements produced resulting from 1 proton-proton bunch collision. Proton Bunch The CMS Detector 40,000,000 Bunch Collisions each second

1 Event generates 1 Megabyte of data 40 Terabytes of data/second!! Requires the storage capacity of 40,000 average home PCs for every second of operation!! Do We Keep it All?!! Need an Electronic trigger to decide which events are important NOT ALL EVENTS ARE INTERESTING

DetectorsMemory Level 1 Trigger Memory Level 2 and 3 Triggers Events Kept for Off-Line Analyses 40,000,000 Events/second generated by CMS detectors Keep Event? - Yes/No? has ~ 3  s (about 120 bunch crossings) to decide Still Keep Event? - Yes/No? Reduced to 100,000 Events/second by Level 1 Trigger Further Reduced to 100 Events/second by Level 2 and 3 Triggers Trigger and Data Acquisition Scheme of CMS 3  s

My Project involved the End Cap Regions of CMS My Project

What are End Caps Anyway? 1 CSC  Rotate 360 o 1 Disk = 1 Station

Combine 4 Stations and Get 1 End Cap End Cap

Profile View of 1 CSC 1 Track Stub Review of CSC Operation

Profile View of 4 CSCs Can construct a full track using track stubs

Real Track Stubs may actually look like this Profile View of o Sector Why Not Always Perfect Lines? Detectors not 100% Efficient Many particles travel through detectors A lot of radiation being produced in throughout the detectors

Other Detector Logic CSC Detector Logic Final Decision Making Logic Keep Event? - Yes or No? Level 1 Trigger Scheme Level 1 Trigger

CSC Detector Logic Scheme CSC Front End Electronics 1 Track Finding Processor Constructs track stubs Constructs Full Tracks Track Stubs Best Tracks CSC Detector Logic

Track Finding Logic Track Sorter Track 1 - ok track - Quality = 5 Track 2 - not very good track - Quality = 1 Track 3 - excellent track - Quality = 10! Track ID = Stubs used to form track Quality = Goodness of track Constructs 6 possible tracks 1 Track Finding Processor Example: IDs and Qualities of 6 tracks Finds Best 3 tracks Best 3 tracks

Best 3 tracks = 3 distinct tracks w/Highest Quality # Must Sort Quality #s to find highest 3 => Need to use Binary Number Sorter Tasks: 1) Sort these Tracks in the Shortest Time Possible 2) Make Sure These Tracks are Distinct Track Sorting Algorithm 6 Track IDs 6 Track Qualities Best 3 Tracks! Track Sorter

How do you Sort Tracks? Use a number sorter to find 3 highest quality tracks Basic Number Sorter: Is A > or = to B ? Input A Input B ‘1’ = yes, ‘0’= no Track Sorter Track 1 has quality 7 Track 2 has quality 9 no = ‘0’ Is 7 > or = to 9 ? => Track 2 is preferred over track 1 Example:

Other Sorting Methods Steps increase as inputs increase For 16 inputs this needs 10 steps For 6 inputs this needs 6 steps Too many steps for 6 inputs There must be a better way Very Bad News Track Sorter Arrange Basic Number Sorter into a tree to sort more numbers

Use VHDL to Sort! Track Sorter What is VHDL? (Very High Speed Integrated Circuits) Hardware Description Language Programming Language like C or Fortran Compiled by computer software like OrCAD to program chips Using VHDL we can create a fast number sorter Works in only 2 steps no matter how many inputs! Completely flexible for any chip type and chip configuration software

How does it Work?? Track Sorter Track A Track B Track C Track D Track E Track F = 21, bit binary number VHDL Number Translator => E has highest quality, D next, and B next Track E Track D Track B All in 1 step Next step List of Track Qualities

Example: Find Highest Value from Input List Track Sorter Each 15-bit value corresponds to a unique ordering of the inputs Number Translator Bits that must be set for Track B to be the best

if ((compa = '1')and(compb = '1')and(compc = '1')and(compd = '1')and(compe = '1')) then inputA = BestTrack; Translate Truth Table to VHDL if-then Statements Truth Table VHDL Track Sorter

Where do the ID #s Come in?? Take the same basic VHDL Sorter and add some disqualification Logic Compare Quality #s Compare ID #s Should Track 1 be disqualified? - Yes/No? Modifies 15-bit number Translates Final 15-bit # into ordering of Inputs Track Sorter Best 3 tracks 15-bit # Track A Qu Track B Qu Track C Qu Track D Qu Track E Qu Track F Qu Track A ID Track B ID Track C ID Track D ID Track E ID Track F ID

Does it Work? Simulations show the sorting method gives correct order Timing Analysis show a 20 ns delay in real chip Finishes within the allowed time Help the performance of Level 1 trigger => Less memory needed to pipeline detector data Less money!

Acknowledgements Kevin Ingersent and Alan Dorsey Professor Darin Acosta - Faculty Mentor S. Ming Wang - Postdoc Raymond Chow - undergraduate assistant Mike Marquez - electrical engineer

Extra Slides for your viewing pleasure!!

1 Track Finding Processor Real Track Stubs may actually look like this Profile View of o Sector How can we possibly construct tracks from this mess!??

1 Track Finding Processor 1234 Look at where a particle passing through stub in station 2 should be when in station 4 and with what parameters Knowing where the particle should be, are any of the stubs in station 4 within the allowed range of values for these parameters? We must extrapolate each track stub to each station

1 Track Finding Processor If yes, then we have successful extrapolation, and connect to stubs to make track After a successful extrapolation, we assign a quality value to the constructed path based on track features

1 Track Finding Processor Quality is based on Momentum and Stations used to extrapolate Higher Transverse Momentum => Less Bending => Higher Quality Top View of 1 Sector

With Some Extra CSCs We can Achieve Complete Angular Coverage 1 Complete Disk (with complete Inner and Outer Disks) = 1 Station Outer Disk of 10 o CSCs Inner Disk of 20 o CSCs

60 o wedge = 1 sector 1 Sector will contain o CSCs from inner disk, and o CSCs from outer disk