1 LAr high speed optical link studies: the status report on the LOC ASIC 1.The LOC ASIC, an introduction 2.The SOS technology 3.LOC1 test results 4.LOC2

1

LAr high speed optical link studies:

the status report on the LOC ASIC

1. The LOC ASIC, an introduction 2. The SOS technology3. LOC1 test results4. LOC2 design status5. Summary

Datao Gong, Andy Liu, Annie Xiang and Jingbo Ye Department of physics, SMU

J. YeDept. of Phys.

Status Report on the LOC ASIC at LAr-Tile-L1Calo upgrade workshop, 11.14.08, CERN

2

The LOC ASIC, an introduction The LOC (link-on-chip) was proposed as a serializer ASIC for the

ATLAS LAr readout upgrade, under the US-ATLAS upgrade program.

The initial idea was to integrate “everything” into one chip, including the optical interface. Fiber would be coupled directly to the chip to spare high speed copper traces on the PCB.

The project started with the SOS technology evaluation in 2006. A first prototype, LOC1 was designed with collaborative effort

between the EE and physics departments at SMU 2007. This prototype provided valuable information on key components, especially the PLL and the serializer structure, for the LOC2 design.

The second prototype, LOC2 is being designed as a 5 Gbps serializer ASIC. This design takes advantage of the CERN common project, the Versatile Link, and uses it as its optical interface. The submission is April 2009.

The whole project benefits tremendously from the CERN GOL ASIC design. We would like to express our gratefulness to many people in the CERN microelectronics group, especially to Paulo Moreira for his very kind help in the LOC project.

J. YeDept. of Phys.


3

The SOS technology A 0.25 micron Silicon on Sapphire

commercial CMOS technology has been chosen for the LOC ASIC development.

A dedicated test chip with transistors, ring oscillators and shift registers has been designed and fabricated for irradiation tests.

Some results from the irradiation tests have been published at RADECS 2007.

The TID test results on transistors are summarized here. The substrate is grounded. Almost no leakage current change; A small threshold voltage change

happens at the very beginning of the irradiation and then remains unchanged with the increase of total dose.

The technology evaluation continues with more detailed studies, supported by the ADR (US DOE) program.

1.0×10-5

1.0×10-5

5.0×10-6

5.0×10-6

ΔV

TH(V

)

0.2

0.1

0.1

0.2

ΔI L

EAK(A

)

NMOS

PMOS

ΔVTH

ΔILEAK

J. YeDept. of Phys.


4

The LOC1 test results

PLL + clock unit

5:1 DFF based serializer


2:1 mux



2:1 mux

8B/10B encoder

16 bit data 2:1 mux

Output driver

Ref. clk

Current driver to VCSEL

CML driver

to VCSEL

CML signal

Control + configuration

Solid line box: implemented; dashed line box: implemented in FPGA.

Eye diagram of an 27-1 pseudo random input data. The data rate is 2.5 Gbps. Large DJ is observed, understood and will be corrected in LOC2

The bit error rate bathtub curve at 2.5 Gbps, the best BER reached is ~10-11.

J. YeDept. of Phys.


5

LOC2 Block diagram and design considerations After many discussions with people in ATLAS Inner Tracker and LAr, and with

more experience of the SOS technology, we now design the LOC2 ASIC in two parts: the 5 Gbps 16:1 serializer and the user interface.

Since the speed of the serializer has been pushed higher and higher, towards the technology limit, a 16:1 structure simplifies the implementation of the high speed circuits.

We move the framing unit into the interface part for better integration with both ATLAS Tracker and LAr readout systems.

We take advantage of the CERN common project, the Versatile Link, and move the optical interface into the VL, so LOC will only provide a CML electrical output.

Interface:Input buffer +

64B/66B + scrambler or 8B/10B

+ output buffer

user data

system clk clkLOC:

16:1 serializationCML output

16

Versatile Link fiber

5 Gbps

The LOC2 prototype aim for 2009.4 submission

Config/control

J. YeDept. of Phys.


6

Interface to different users

For ATLAS Tracker, the input to the optical link may contain DC balance coding that may need to be removed to save on bandwidth overhead.

For LAr, link bandwidth is the premium. We propose the interface chip/function block to be:

user data

system clk

clk

16Input buffer: Extract user data and change width to 64B or 8B

64B/66B+ scrambler or 8B/10B

Output buffer:Change data to 16B, LVDS

Divided by NLow speed PLL + clock fan-out

J. YeDept. of Phys.


7

The 16:1 serializer at 5 Gbps

The serializer is based on a cascade of 2:1 multiplexing units with only the last stage working at 2.5 GHz clock. The logic structure is much simpler than the traditional shift register based (20:1) serializer in which all registers work at 2.5 GHz clock.

The Critical components are:1. The 2.5 GHz PLL with low jitter and a 50% duty cycle output.2. The static D-flip-flop (chosen for SEE immunity) and the 2:1 MUX. The last stage is

specially designed to work with the CML driver. 3. The CML driver.

We have finished schematics level simulations. Layout and simulations are in progress.

2.5GHz PLL + clock fan-out

625M312.5M 1.25G 2.5G

16bitLVDS

312.5MHzRef Clk

5Gbps

CM

L D

river

LVD

S-T

o-C

MO

S

D flip-flop

2:1MUX

J. YeDept. of Phys.


8

LOC2 design status Speed comparison of 0.25 μm SOS and BulkCMOS

(TSMC) with inverter, and adjust the PMOS/NMOS transistors ratio for the same 01 and 10 transition time. done.

Choose a static D-flip-flop design that meets the 5 Gbps speed requirement. done.

Complete schematic level study on the 16:1 serializer. in progress.

Layout and verification of the 16:1 serializer. in progress.

Interface. may need help in manpower or postpone it into FPGA for the moment.

J. YeDept. of Phys.


9

Summary The LOC ASIC proposal evolves with time. We incorporate into

our LOC design the development from the Versatile Link project and decide to move the optical interface from the LOC to the V.Link. The LOC now is a 16:1 serializer plus user interface. Different interface ASICs or function blocks will be developed according to the application of the LOC.

Technology evaluation on the 0.25 micron SOS technology produced encouraging positive results and enables us to go ahead with the LOC design using this technology. More studies will be performed on this technology with support from the ADR program.

The design work for the present prototype, LOC2, is in progress. Simulation on critical components indicate that 5 Gbps serial data rate is hopeful.

We aim for the April 09 submission, and the tests in lab July 09. We will provide demo-link and system design document for groups that are interested in using this chip in the fall of 2009.

J. YeDept. of Phys.


10

Backup slides

J. YeDept. of Phys.


11

The inverter PMOS/NMOS ratio adjusted to

have the same 10 and 01 delays.The ratio is: n*(1.9/1.4) where n = 1,2,3,4…

Basic layout, multi-finer layout checked to optimize speed. The delay is about 32~35 ps (drive itself), corresponding to a frequency of about 30 GHz. Agree with the manufacturer's tech notes, and comparable with speeds achieved in 0.13 to 0.15 micron bulk CMOS technology.

A comparison is made with 0.25 micron bulk CMOS (TSMC) on the same inverter design. Simulation shows a 60 ps delay with the same layout and driving condition.

schematics

layout.

J. YeDept. of Phys.


12

The D-flip-flop (DFF) We started out with the C2MOS type of DFF used in GOL, but moved to the TGDFF:

~20% faster, and at least the same SEE immunity (Ramanarayanan, Upenn). Different transistor size, single finger and multi-finger layouts are checked. The total

delay is 292 ps (slowest or the S-S corner). This indicates a 5 Gbps serializer possible, because the time needed for a basic unit (DFF+mux) is 400 ps.

schematics

Mostly single-finger layout

multi-finger layout

J. YeDept. of Phys.


13

Test results on LOC1

The in-lab tests: Electric output: rise/fall times, amplitude. Eye diagram. Understand the jitter. Bit error rate.

SEE with 200 MeV protonFuture tests planned on LOC1

J. YeDept. of Phys.


14

LOC1 block diagram

Five function blocks: the input 8B/10B encoders, the 20:1 serializer, the output driver, the PLL and clock unit, and the control and configuration.

Solid line box: implemented; dashed line box: not implemented. Loc1 is the 1st prototype only good for tests in lab.

PLL + clock unit


5:1 DFF based serializer2:1 mux


5:1 DFF based serializer2:1 mux

8B/10B encoder

16 bit data 2:1 mux

Output driver

Ref. clk

Current driver to VCSEL

CML driver

to VCSEL

CML signal

Control + configuration

Inpu

t

Output

8B/10B encoder

J. YeDept. of Phys.


15

LOC1 basic electrical characteristics Input power supply: 2.5V (Analog/Digital), 3.3V VCSEL Power consumption (with current configuration)

~200mW. Input data signals LVCMOS 2.5V

Reference clock 62.5MHz 20bit 8B/10B encoded input data

J. YeDept. of Phys.


16

LOC1 test system diagram

• Tektronix 20GHz real-time oscilloscope (DSA72004) and Anritsu 12G BERT are used in the tests.

• The VCSEL driver is not functioning. All tests are carried out through the CML driver on electrical signals only.

• Tests performed at the test point (see figure) are the CML signal amplitude, the rise and fall times, and the jitter measurement and analysis.

• The Bit Error Rate (BER) is measured with the Anritsu BERT.

ElectricalDifferential

Data

Clock

LVDSDriver Board

LVDSReceiver

Board

Clock distribution

I/O to PC

FPGA:PRBS generator8B/10B encoderError detection

LVDSInterfaceBoard

LOC1Carrier Board

TLKCarrier Board

FPGA Board

TLKInterface Board

Test point

Data

Clock

J. YeDept. of Phys.


17

The rise and fall times and the amplitude

• The logic in the digital part, and the PLL + clock unit are functioning as designed. The output serial bits are correct, checked on the scope.

• The differential signal amplitude is about 240mV, measured by a differential probe. This is much lower than the 400 mV design spec.

• The rise and fall times are data pattern dependent. They are measured to be around 150ps. This is boarder line for 2.5 Gbps signal.

• The reason for the above problems are not completely known. The whole driver design will be abandoned. We will use a new driver design in LOC2.

Comma (k28.5) code on differential electrical channel

J. YeDept. of Phys.


18

Eye diagram

Eye diagram of the Input 27-1 Pseudo random data.UI = 400ps, for 2.5 Gbps.

• An eye diagram is measured.• There is huge jitter generated

inside LOC1.• Careful measurements were

carried out to identify the components as well as possible sources of this jitter.

• We think that we understand this problem very thoroughly. The huge DJ comes from the 4-arm 2-stage mux serializer design. This architecture is not necessary. New serializer design will be implemented in LOC2 to address this problem.

• The RJ comes mostly from the PLL. Efforts are spent to improve the PLL design for LOC2.

J. YeDept. of Phys.


19

Jitter measurement method

DSA72004 employs the software package TDSJIT3 for jitter analysis.

Sample is recorded at 40ps and up to 50M data points in one data file — a snapshot of 2ms waveform length, for long term jitter analysis.

Algorithm performed in the frequency domain.

RJ and DJ components are extracted.

J. YeDept. of Phys.


20

Jitter measurement with TDSJIT3

Jitter spectrum Results on jitter in the serial data bit stream:• RJ_rms = 10ps, DJ_pk-pk = 230ps. • Out of the 230 ps DJ, the periodic jitter contributed by the

serializer structure is175ps. This can be eliminated in LOC2.

• Total jitter (@10-12 BER) = 310ps. Reminder: UI = 400ps.

J. YeDept. of Phys.


21

The Periodic jitter

• A large periodic jitter is observed. • This strong data depending, n×492KHz jitter structure

(2.5GHz/127/40 = 492KHz) is a clear indication that the jitter comes from an unmatched 4-arm 2-stage serializer.

• New serializer structure will be implemented in LOC2 to address this problem.

J. YeDept. of Phys.


22

Bit error rate measurement

BERT signal generator 62.5MHz FPGA board

LOC1BERT Error Detect

62.5MHz 20bit Data

Serialized Data

2.5GHz

BER

The BER measurement system

We use the Anritsu MP1763/1764 BER Tester for this test

The input amplitude is 110mV because of the Differential to single-ended converter.

J. YeDept. of Phys.


23

The BER bathtub curve

The bathtub curve at 2.5Gbps, the best BER reached is ~10-11.

• The bathtub curve is obtained by shifting the reference clock with respect to the parallel data.

• From this curve we extract: DJ_pk-pk = 235ps, RJ_rms =18ps, and total jitter = 478ps (@10-12 BER)

• The low signal amplitude might cause the large RJ measurement: The BERT error detect requires 250 mV amplitude input, but our input is 240mV/2 ~ 110mV.

J. YeDept. of Phys.


24

LOC1 SEE test with 200 MeV proton (very preliminary)

LOC1+TLK2500 link operates at 2.5 Gbps with the LOC1 exposed in 200 MeV proton beam.

Flux: 5×106 p/cm2/sec. Error free for 30 minutes, a fluence of 9×109 p/cm2. After that TID effect may kicked in. Data analysis is on-going and more tests will be needed.

SEE cross section less than 1×10-10 cm2/proton. Or less than 1 error per link every hour at SLHC.

J. YeDept. of Phys.


25

Previous LOC2 block diagram

The preliminary LOC2 block diagram

16 bit Input registerLVDS to LVCMOS

2:1 MUX to 8 bit

8B/10BComma

27-1 PRBS PLL + clks

MU

X

Cntrl config

Odd bits shift register

Even bits shift register

Latch

CML driver

2:1

MU

X

Data

Clk_ref

Cntrl/Config

16 LVDS 10 bit

Serial output to Versatile Link

Inpu

t

Output

LOC VL fiberParallel data GBTx VL fiber

Parallel data

System implementation

Documents

1 LAr high speed optical link studies: the status report on the LOC ASIC 1.The LOC ASIC, an introduction 2.The SOS technology 3.LOC1 test results 4.LOC2