ATLAS Level–1MuonBarrel RODbus Preliminary Design Revie...A. Aloisio ATLAS Level–1MuonBarrel...

A. Aloisio

ATLASLevel–1 Muon Barrel

RODbusPreliminary Design Review

Mar.12, 2002Alberto AloisioINFN - Sezione di Napoli, Italy

e-mail: aloisio@na.infn.it

Mar. 12, 2002 2

A. Aloisio

Overview

� Muon Level-1 Trigger and Data path� ROD crate architecture� RODbus features� Measure, models & simulations� Test results� Conclusions

Mar. 12, 2002 3

A. Aloisio

From PAD to ROD/SL

ROD

TriggerSL

MTCPI

RXData

PAD Logic

� Optical out from PAD to:– SL– ROD

� 12 bit @ 40MHz on the trigger path; synch, fixedand low latency link required to guarantee timing.

� 16 bit + strobe on the data path.

Mar. 12, 2002 4

A. Aloisio

ROD Crate layout

Mar. 12, 2002 5

A. Aloisio

RODbus requirements

4848+48

48+48 4848

48

ROD RX

RODbus (SL)

SL

MTCPI

SL

MTCPI RODbus (RX)

RX12 16

� SL to ROD– 48 bit@40MHz– synchronous with low

latency

� RX to ROD– (16+8 bits x 2 = 48 bits)

� ROD to RX/SL(not shown)– Low skew, low jitter

clock distribution– control signals

(RST*, LVL1A, ...)

Mar. 12, 2002 6

A. Aloisio

Strategy

� Common platform for both SLs and RXs to RODbackbones:– SL asks for a synchronous design.

– Low skew, low jitter signals need a differentialsignaling scheme.

– Doubling data lines saturates the pin availability (andboard space) if a SerDes approach is not used.

– Low voltage, low power (and high performances)suggest LVDS.

– System level “non-timing” signals (as RST*) can runTTL.

Mar. 12, 2002 7

A. Aloisio

Choosing the chip-set

� DS90CR483/4 chip-set, a 48:8 bit + clock LVDSpseudo SerDes

� Deskew and Preemphasis for improved performances

� Data transfer rate @ 7x master clock frequency

backplane

Mar. 12, 2002 8

A. Aloisio

Partitioning the problem

� To optimize board layout and backplaneperformances, we decided to assign the SLsto ROD and the RXs to ROD channels totwo “add on” backplanes.

� In this review, we present the prototype ofthe RXs to ROD backplane, assigned to J2

� The SLs to ROD backplane will be doneusing the same technology, mapped to J0

Mar. 12, 2002 9

A. Aloisio

RODbus: RXs to ROD

� 3 slot backplane to link2 RXs to one ROD

� 10 layer stack-up

� Diff. microstrip forLVDS pairs and singleended for TTL lines onseparate planes

� Plug-in to the VME64rear side. TTL linesterminated as VME

Mar. 12, 2002 10

A. Aloisio

ROD slot pinout

� ROD distributes clock and 8 system control signals

� Receives 48bits@40MHz from each adjacent RX

TTL lines

RXB

SerDes

RXA

SerDes

clock

Mar. 12, 2002 11

A. Aloisio

Physical layout

� Diff. LVDS pairsrouted as edgecoupled microstrips

� All pairs have thesame length (10 miltolerance).

� Noisy TTL lines arerouted on a separateplane.

RXB clock

LVDS diff.

RXA clock

LVDS diff.

SerDes clock

LVDS diff.

SerDes clock

LVDS diff.

Mar. 12, 2002 12

A. Aloisio

TDR interface

� Two test fixtures havebeen designed to interfacethe RODbus to TDR

� Diff. Impedance profilecan be measured

� Impact of connectors,vias, stubs and holes(present in the realenvironment) can beevaluated

Mar. 12, 2002 13

A. Aloisio

TDR profile

� TDR step probesbackplane line andconnectors

� PCB stack-up is tested

� line length andimpedance are measured

� lumped L/C (connectors,vias, solder pads) arevisible

50 Ohm

coax cable

test

fixture

backplane

connectors

L=575ps

Open load

reflection

50 Ohm

LVDS line

Mar. 12, 2002 14

A. Aloisio

Testing the SerDes: ROD slot

� ROD emulation:– Distributes Clock to RXs (from ext. or local source)

– Receives serial streams from RXA and RXB

Interface to

ParBERT

Ext. Clock

Deserializer

(RXB)

Clock

LVDS driver

Deserializer

(RXA)Local Clock

Mar. 12, 2002 15

A. Aloisio

Testing the SerDes: RX slot

� RX emulation:– Receives Clock from ROD

– Transmits serial streams to ROD

Interface to

ParBERT

Clock

LVDS receiver

Serializer

Mar. 12, 2002 16

A. Aloisio

TDR system test

RODbus

VME

J2/P2

RXB

Emulator

ROD

EmulatorTDR step

launch point

� The entire RX-to-ROD connection is analyzed

� Unpopulated RX and ROD emulators are used toevaluate the signal integrity using TDR

Mar. 12, 2002 17

A. Aloisio

Impedance profile

� The TDR reflectedstep shows thediscontinuities (vias,holes, connectors, ...)

� The impedanceprofile is deriveddeconvolving themultiple reflections

Mar. 12, 2002 18

A. Aloisio

Modelling the system

� A PSpice model hasbeen developed, basedon the impedanceprofile

� The model is based onideal (lossless) linesegments

� Model validation is performed by simulatingthe TDR reflected step. Results are in excellentagreement with experimental data

Mar. 12, 2002 19

A. Aloisio

IBIS models

Typical output stage

Typical input stage

� IBIS is an ANSI standard tomodel I/O buffers mainly forsignal integrity issues

� Different from SPICE, IBISgives only a behavioraldescription - disclosing noproprietary circuit information

� The I/O buffer is characterizedby means of a standard templateof VI curves and stray L,C and R

Mar. 12, 2002 20

A. Aloisio

DS90CR483 LVDS Output

Mar. 12, 2002 21

A. Aloisio

System simulation

TX LVDS

PSpice models derived

from IBIS

RODbus

PSpice model

RX LVDS

PSpice models derived

from IBIS

� PSpice system simulation includes backplanes,connectors, stubs, SerDes TX and RX

Mar. 12, 2002 22

A. Aloisio

Simulation caveat

� Interconnection Model is based on idealtransmission lines

� Crosstalk, ground bounce, probe loading arenot modeled

� TDR-derived models are less accurate thelonger the step travels

� SerDes IBIS model only specs typ. values,pre-emphasis and deskew not modeled

� Standing wave needs long simulation runs(>15h to simulate 2us on a Pentium II – 350)

Mar. 12, 2002 23

A. Aloisio

SerDes LVDS clock - TX side

LVDS SerDes

Clock @ TX

simulation

Mar. 12, 2002 24

A. Aloisio

SerDes LVDS clock – RX side

LVDS SerDes

Clock @ RX

simulation

Mar. 12, 2002 25

A. Aloisio

Test results (preliminary)

� Tests with 2^15 PRBS show no errorsin the SerDes operations up to 60 MHz(420 Mbit/s). Fine tuning should allowreaching 80 MHz (560 Mbit/s).

� Latency is 5 clock cycles (125ns @ 40MHz), dominated by the SerDes specs.RODbus Tpd (570 ps) is negligible.

Mar. 12, 2002 26

A. Aloisio

More tests to be done

� RODbus performances in a real VME(noisy) environment

� EMI

� BER vs. clock frequency and differentpatterns

� 48-bit 2^16 Pseudo Random WordSequence (PRWS) test requires acustom platform (generator, analyzer,adapters, …) to be developed

Mar. 12, 2002 27

A. Aloisio

Conclusions (1/2)

� RODbus is the proposed backbone for all theROD crate interconnection needs

� It specifies the physical layer (backplane, lineimpedance, levels) and the logical layers(SerDes specs)

� Data transfer rate is 7x the master clockfrequency (280 Mbit/s @ 40MHz) requiringcareful PCB layout for backplane and users’boards. Emulators have been designed proof ofconcept, tests and PCB layout reference

Mar. 12, 2002 28

A. Aloisio

Conclusions (2/2)

� RODbus has been in-deep characterized byusing TDR techniques. SPICE models have beenderived and validated. SerDes SPICE modelshave been crafted starting from IBIS files.

� The entire system has been simulated withPSpice and results show good agreement withexperimental data.

� Analog simulations has been successfully usedto evaluate the system performances and signalintegrity issues.