A Study on Full Digital Clock Data Recovery (CDR) CDR Fundamentals Preamble bits Timing get sample...

30
1 Philipus Oh, Win Chaivipas, Akira Matsuzawa Tokyo Institute of Technology Graduate School of Science and Engineering Department of Physical Electronics Matsuzawa Laboratory A Study on Full Digital Clock Data Recovery (CDR) Matsuzawa Lab. Tokyo Institute of Technology

Transcript of A Study on Full Digital Clock Data Recovery (CDR) CDR Fundamentals Preamble bits Timing get sample...

1

Philipus Oh, Win Chaivipas, Akira Matsuzawa

Tokyo Institute of TechnologyGraduate School of Science and Engineering

Department of Physical ElectronicsMatsuzawa Laboratory

A Study on Full Digital Clock Data Recovery (CDR)

Matsuzawa Lab.Tokyo Institute of Technology

2

Contents

• CDR fundamentals • PLL based CDR and Oversampling CDR • Proposed CDR concept • Concept verification by system level

simulation• Merit & Demerit• Summary and Future work

Matsuzawa Lab.Tokyo Institute of Technology

3

CDR Fundamentals

Preamblebits Timing

get sample Incoming Data

A part of receiver that:

QD

pre-amplifier

Clock DataRecovery

Data out

VTT

Datain

Receiver

QD

-VTT

QD QD

V

Preamble bits (to make sure CLK information is obtained before sampling the data)

Data

Matsuzawa Lab.Tokyo Institute of Technology

4

Applications

Any digital communication systems/network that use serial link:• Optical Receivers (SONET)• Chip to chip• Peripheral I/O• Crossbar switches for Asynchronous Transfer Mode

Matsuzawa Lab.Tokyo Institute of Technology

5

CDR FundamentalsData Jitter Sampling CLK

Jitter

1 bit

The optimum point to sample data is at the middle of the bit window (black arrow)

Bit error only occurs when the CLK jitter is greater than 0.5bit period

Matsuzawa Lab.Tokyo Institute of Technology

6

CDR classification

• Phase Locked Loop• Oversampling Clock Recovery

Phase Locked Loop:

Low passfilter

Phase Detector

&Charge pump

VCO

DCLK

Q

Data RecoveredClock

Matsuzawa Lab.Tokyo Institute of Technology

7

PLL based CDR

Data

CLK

Y

B

A

X

Data

CLK

Y

B

A

X

Phase detector in PLL for CDR must be able to detect edges and phase difference in random data.

Classified into:• Hogge’s Analog Phase Detector• Alexander Binary Phase Detector

Hogge’s Phase detector Timing Waveform Transfer Function

The difference between X & Y will show the phase difference amount and polarity Matsuzawa Lab.Tokyo Institute of Technology

8

PLL based CDR

Y

XD Q D QData

CLK

D Q D Q

Q1

Q2

Q3

Q4

X

Y

Alexander Binary PD Timing Waveform Transfer Function

When data edge leads

When CLK edge leads

The result of phase detection only tells the polarity,whether leading or lagging.

Thus, jitter is constantly produced at every phase detectionThe amount of phase difference is not detected

Matsuzawa Lab.Tokyo Institute of Technology

9

PLL Issue

• Low frequency acquisition, within 25% of VCO center frequency

• Long lock time, about 100 preamble data(depends on bandwidth, which also affect jitter performance, there’s a trade off)

• Possibility of harmonic lock

Matsuzawa Lab.Tokyo Institute of Technology

10

Oversampling CDR

Data is obtained by sampling data by multiphase clock:

Data is processed to choose which values represent the data bit:

Matsuzawa Lab.Tokyo Institute of Technology

0 11 011 00 0 1

11

Oversampling CDR Issue

• At least 3 sample is needed for each data bit and special clock is needed for sampling.(When input frequency change,

sampling clock must also be changed.)

• Jitter due to phase quantization

Matsuzawa Lab.Tokyo Institute of Technology

12

Proposed CDR Intro

The proposed CDR is maintained to sample around the middle of the bit window (colourful arrows)

But NEVER at the middle of bit window!!

Data Jitter Sampling CLKJitter

1 bit

Matsuzawa Lab.Tokyo Institute of Technology

13

TDCData

X=Measured data period

Data period =1001.3 psTDC resolution =10 psExample:

Proposed CDR

Due to the fact that △Ф=△ω.T, if we use X directly,large BER will occur:

= sampling using X as sampling period

=sampling using ideal data period

=sampling error

X=1000 ps or 1010 ps

Matsuzawa Lab.Tokyo Institute of Technology

14

Alternate Sampling

= 1010 ps (Upper boundary)

= 1000 ps (Lower boundary)

If we use this sampling period alternately :

Data period = 1001.3 psTDC resolution = 10 ps

Error will not occur !

Data period lies within 1000 ps and 1010 ps(TDC step interval)

= ideal sampling point= sampling point using result from TDC directly

Matsuzawa Lab.Tokyo Institute of Technology

15

Data period =1001.3 psTDC resolution =10 ps

X =1000 ps

A = X + 2*TDC resolution= 1020 ps

B = X + TDC resolution= 1010 ps

C = X = 1000 ps

D = X - TDC resolution= 990 ps

E = X - 2*TDC resolution= 980 ps Lower Boundary =1000 psUpper Boundary =1010 ps

A2>A1B2>B1C2<C1D2<D1E2<E1

Compare

Obtaining Upper and Lower boundary

Theoretically, only 3 preamble data is needed!! Matsuzawa Lab.Tokyo Institute of Technology

X

TDCData

X=Measured data period

Edge data 1 A1B1C1D1E1 Edge data 2

A2B2C2D2E2

16

Sampling Based on Feedback systemCompare distance between data edge and sampling edge (Y’s) with X/2

Comparison is executed at every data edge

Y1

Y2

1000psec

Y3

Y4

Y5

Y6

Data in Data out

Is Y>=X/2?

Upper boundary

Lower boundaryYes

No

X/2Y

= 1010 ps (Upper boundary)

= 1000 ps (Lower boundary)= ideal sampling point

X=Measured data period

Matsuzawa Lab.Tokyo Institute of Technology

17

TDCDFF

TDCProgrammable

Counterfor Sampling

Incoming Data

Recovered data

Simulation Result in Verilog-A

Simulation environment:• PRBS 27-1 data• 350 bits• TDC resolution 10 ps• Tdata 1001.3 ps

Matsuzawa Lab.Tokyo Institute of Technology

18

Y1

Y2

1000psec

Y3

Y4

Y5

Y6

Simulation Result in Verilog-AY(psec)

Time (nsec)

Y value is maintained around halfbit period (500.65 ps)

ideal500.65ps

Matsuzawa Lab.Tokyo Institute of Technology

19

Deterministic Jitter estimationThe peak to peak jitter is dependent upon TDC resolution and data type:

2N-1 N*TDC resolution The RMS jitter is dependent upon the time distance between data period and lower/upper boundary

Deterministic Jitter Histogram

-5

0

5

10

15

20

-20 -10 0 10 20 30 40 50 60

Jitter (ps)

Event TDC resolution 10ps

Data period 1001.3ps

Matsuzawa Lab.Tokyo Institute of Technology

20

Realization issues

• Counter with tens of ps speed is difficult• Counter is replaced by multiphase clock

generator and logic• [5] has shown 10 ps multiphase clock is

possible in TSMC 0.18 μm

[5]“A Sub-10-ps Multiphase Sampling System Using Redundancy” Lee. Li, Weinlander. D, Yang. C-K K

Matsuzawa Lab.Tokyo Institute of Technology

21

Clock Recovery Stage

Matsuzawa Lab.Tokyo Institute of Technology

22

Data Recovery Stage

Matsuzawa Lab.Tokyo Institute of Technology

23

With logic and ideal Multiphase clockY value

Less than 5 preamble data was used

Matsuzawa Lab.Tokyo Institute of Technology

24

Merit and demerit

MERIT• Fast lock• Infinite frequency

acquisition range• Robust• Scalable

DEMERIT• Jitter occurs in every data sampling

(deterministic jitter)• Jitter depends on data type

Deterministic Jitter is dependent on digital resolution. As technology scales, jitter will reduce as well

25

Comparison to PLL

PLL based CDR Proposed CDRLock Time > 100 preamble data.

Depends on loopbandwidth

A few data bits

Jitter Depends on analog block’s noise, loop bandwidth, mismatch & variations

Depends on TDC resolution & multiphase clock jitter

Freq acquisition

Finite Depends on TDC memory capacity, theoretically infinite

Scalability NO YES

26

Comparison to OversamplingOversampling CDR Proposed CDR

Lock Time Within a few data bits Within a few data bits

Jitter Depends on multiphase clock jitter

Depends on TDC resolution & multiphase clock jitter

Response to Change ofInput freq

Take time for the multiphase clock to relock

Fast, only a few data bits

Scalability YES YES

27

Comparison

PRBS 28-1 P-P Jitter 280 ps

[1] “An Auto-Ranging 50-210Mb/s Clock Recovery Circuit with Time-to-Digital Converter”. ISSCC’99, Seoul National University

Use 3 analog loop, lock time takes several μsec

Matsuzawa Lab.Tokyo Institute of Technology

28

Summary and Features

• A new concept for Full digital CDR has been verified in verilog-A

• The proposed CDR has infinite frequency acquisition.

• The highest operating frequency is only limited to speed of digital circuit.

• Scalable, and as device shrinks, jitter and speed are automatically improved

• To sum up, fast lock, mobility, robustness, and scalability is this circuit’s main attraction

Matsuzawa Lab.Tokyo Institute of Technology

29

Future Works

• Phase detection and Phase compensation circuit

• Improvements against data jitter (Jitter tolerance)

Matsuzawa Lab.Tokyo Institute of Technology

30

Reference

[1] “An Auto-Ranging 50-210Mb/s Clock Recovery Circuit with Time-to-Digital Converter”. Park, Joonbae; Kim, Wonchan[2] “Phase Interpolator Using Delay Locked Loop”. Kim, Taesung, Kim, Beomsup[3] “A 0.5um CMOS 4Gb/s serial link transceiver with data recovery using oversampling.” C.-K. K. Yang, R. Farjad-Rad, M. Horowitz.[4] “A Multibitrate Burst-Mode CDR Circuit With Bit-Rate Discrimination Function From 52-1244Mb/s.” S. Kobayashi, M. Hashimoto[5] “A Sub-10-ps Multiphase Sampling System Using Redundancy” Lee. Li, Weinlander. D, Yang. C-K K

Matsuzawa Lab.Tokyo Institute of Technology