R3 for ABCN Discussion

2R3 for ABCN Discussion17 Feb 2011

Regional Readout• Level-1 Muon/Calo system identifies a Region of Interest (RoI)• RoI mapped to set of front-end modules• Regional Readout Request (R3) signal is sent ONLY to modules

in RoI• Modules readout Regional Data (RD) in a similar manner to

normal data• Regional data in intercepted on the ROD, forwarded to the track

finder• Track finder contributes to Level-1 Trigger decision

ROD Crate x20

ROD Crate x20

RODsRoI MapperL1

Muon/Calo StavesRoIs

Track FinderL1

3R3 for ABCN Discussion17 Feb 2011

USA15

ROD-Crate

RoI/R3 to Stave Flow1. Muon and/or Calo triggers2. RoI Mapper

• - sync/align (Unclear on RoI data format and rate from L1)• RoI Mapper fanout

3. GBT to ROD Crate timing interface (TIM)• Unclear GBT data latency

4. ROD-crate• TIM to ROD, with possible decoding (sent over backplane)(50ns)• ROD decode, map, routes to each channel(25ns)

5. GBT to SMC• Unclear GBT fast command and data latency• 100m fibre to FE

Muon Trigger

RoI MapperROD

Calo Trigger

SMSMSMSMStave

SMSMSMSMStaveROD

SMC

SMC

1700ns 150ns 100ns 75ns 600ns

Group Fanout TIM

GB

T

GBT

4R3 for ABCN Discussion

Working parameters

Although large buffers provide for long latencies, it is beneficial to be efficient:

• More time for trigger decision

• Smaller FE buffers

• Less R3 induced dead-time

Latency can be reduced by:

• Fast R3 distribution (as is reasonable)

• Compress R3-Data to reduce packet count

• Prioritised R3-Data packet handling

17 Feb 2011

5R3 for ABCN Discussion

R3-Data details•Leaves chip ASAP

–Queue jump

•Data reduction–E.g. 3 clusters, no width–Cut on cluster width (>3 = discarded)–Total nit/cluster count in packet–1 packet only (or 2, only)

•Easily identifiable packet–Quick routing on ROD

17 Feb 2011

6R3 for ABCN Discussion

Pipeline

L0 Buffer

L0

R3

R3 Data Format

L1 (L0ID)

L1 Data Format

Strips Readout

Synchronous R3

L1 Derandomise

+ minimal processing on the ABCN

+ simple test environment

+ low latency- Off-detector R3 distribution needs to be fast

- Deadtime- More bandwidth for R3 on the stave (more lines, faster)17 Feb 2011

7R3 for ABCN Discussion

Yes/No R3

+ Asynchronous

+ Simple of buffer management

- higher bandwidth than Yes only option

17 Feb 2011

L0 BufferL0

R3Y

L1 (L0ID)

L1 Data Format

Strips Readout

L1 Derandomise

R3 Derandomise

R3 Data Format

Pipeline

R3N

RD FIFO

8R3 for ABCN Discussion

Yes only (with L0ID) R3

See Mitch’s diagram ....

+ very flexible

- longest latency

17 Feb 2011

9R3 for ABCN Discussion

BoundariesUseful to setup some working limits (even if they get changed in

AUW)• L0 rate - 500kHz• L1 rate - 50kHz• R3 data volume (simple latency)

–1 packet per R3 per ABCN

• R3 overall latency (queuing time etc)–Max readout time – 50us?

• R3 Data compression strategies/parameters–Report number of hits (or clusters?) - up to 32?

• Or maybe a course hitmap (1 bit/8)

–Only process clusters <4 strips wide–Only report first 3 clusters

17 Feb 2011

10

R3 for ABCN Discussion

Some More Thoughts ...• Packet inspection?

–May improve intra-chip prioritising

• R3 deadtime?–Dropped packets affects trigger efficiency–Track finder forces L1 for missing R3-Data events– rate reduction

• Slow chip – 40MHz - only do 80MHz at output–DDR?–Timing issues?

17 Feb 2011