TITLE. In-depth Analysis of DDR3/DDR4 Channel with Active Termination. Image. Topic: Topic: Changwook Yoon, (Intel)

Transcription

1 TITLE Topic: o Nam elementum commodo mattis. Pellentesque In-depth Analysis of DDR3/DDR4 Channel with Active Termination malesuada blandit euismod. Topic: Changwook Yoon, (Intel) o Nam elementum commodo mattis. Pellentesque malesuada blandit euismod. Nam elementum commodo mattis. Pellentesque Woojin o Lee, Jack Chui, Gawon kim, Dharmesh Bhakta (Intel) Dan Oh (Samsung Electronics) malesuada blandit euismod. Image Topic: o Nam elementum commodo mattis. Pellentesque malesuada blandit euismod.

2 Age da Introduction Basic of On-die termination Comparison of on-die termination: Passive/Active Non-Linearity in Active Termination I-V curve in active termination Impacts Impact of Non-Linearity in Active Termination SSN variation Timing variation Summary

3 Basic of Active Termi atio SSN Z T1 I T TX Channel Channel Z T2 Active termination is hard to meet desirable impedance Unexpected impedance determines signal quality in o DC level o Reflection noise o Crosstalk o SSN

4 Compariso of Active Termi atio Passive ODT Active ODT Passive Active Layer Poly PMOS Area Large Small PVT variation Less-sensitive sensitive I-V Linearity Good Bad

5 I-V Curve in Active & Passive Termination Middle C P =0.5 V CC V CC I out I out O P =C P V CC V out V out Low C P =0~0.49 High C P =0.51~1 Passive termination: current is inversely linear to output voltage Active termination: current is NOT linear to output voltage Calibration point is an operating voltage at desirable impedance

6 I-V Curve & Impedance 40Ω 60Ω 90Ω C P =0.35 O P =420mV As R T is getting less, I OUT becomes larger I OUT at Lower V OUT is more stable than Higher V OUT More NON-Linear R T is more constant at higher V OUT Need to measure linearity

7 Non-linearity Coefficient ΔR L C = ΔV X% 90Ω C P +/- 5% -5% O P +5% L C (Ω/V) 40Ω Ω Ω Ω 40Ω At calibration point, voltage range needs to be defined first. Linearity at C P is R T variation within defined voltage variation High LCP means High NON-linearity

8 DC Level 1.2 R S I out RT,V Channel As R T goes higher, low-voltage and current becomes lower (Larger eye-height) As R S goes higher, low-voltage goes higher (smaller eye-height) and current goes lower (less power)

9 Crosstalk from Active Termination B Driver R S,V =34 Victim Aggressor x 6 R T,V 1.2 R T,A A Measured crosstalk at victim depending on R T,A or R T,V Crosstalk is more sensitive to R T,V than R T,A Lower R T (A) has less noise but larger noise variation than higher R T (B)

10 SSN from Active Termination 1.2 Driver R I,V =34 Victim Coupling SSN x 38 R T,V R T,A PDN Measured SSN at victim depending on R T,A or R T,V under Coupling and No Coupling Without coupling, SSN becomes less as R T,A or R T,V goes higher With coupling, noise follows crosstalk trend at R T,V but SSN trend at R T,A

11 Active Impedance Setting V CC 19.1mA R P =40 I out +/- 5% +/- 5% V out R a = V CC Vout I out 7.5mA 0.35VCC needs higher current to make 40ohm than 0.75VCC Linearity (+/-5%) at 0.35VCC is smaller than 0.75VCC C P (0~1) Linearity (Ω/V) Low C P (0.35) 40.8 High C P (0.75) 57.5

12 Voltage (V) Voltage (V) Voltage (V) ISI Waveform Ideal V CC R A =40 Ideal 40Ω Active R at C P =0.35 Active R at C P =0.75 Channel V out C P (0~1) Linearity (Ω/V) V LOW (mv) R TERM (Ω) Jitter (UI) Ideal Low C P (0.35) Time Time Time High C P (0.75) At POD, low-voltage at 0.75xVCC is close to ideal 40 Though linearity is bad, calibration point is more important to get better timing error

13 Voltage (V) Voltage (V) Voltage (V) ISI+Crosstalk Waveform 1.3 Ideal V CC C P (0~1) Victim Aggressor Linearity (Ω/V) IO Noise (mv) V out Jitter (UI) Low C P (0.35) High C P (0.75) V 0.35V Time (ns) Active R at C P =0.35 Active R at C P =0.75 Crosstalk is bigger at 0.75xVCC but jitter is less Time

14 Voltage (V) Voltage (V) ISI+SSN Waveform V CC with PDN Burst Victim V out PDN C P (0~1) Linearity (Ω/V) PWR Noise (mv) IO Noise (mv) Jitter (UI) Low C P (0.35) High C P (0.75) Active R at C P =0.35 Active R at C P =0.75 PWR & IO noise at 0.35xVCC is bigger due to larger current Noise ratio at 0.75xVCC is bigger due to larger R T Time Time

15 Voltage (V) Voltage (V) ISI+Crosstalk+SSN Waveform V CC with PDN PDN Victim Aggressor Burst V out C P (0~1) Linearity (Ω/V) PWR Noise (mv) IO Noise (mv) Jitter (UI) Low C P (0.35) High C P (0.75) Active R at C P =0.35 Active R at C P =0.75 PWR noise at 0.35xVCC is worsen, but IO noise is better Timing error at 0.75xVCC is better despite larger IO noise Time Time

16 Channel Summary C P PWR Noise IO Noise Jitter ISI Xtalk SSN ALL Low C P (0.35) High C P (0.75) Low C P (0.35) High C P (0.75) Low C P (0.35) High C P (0.75) Low C P (0.35) High C P (0.75) High calibration point is better jitter than low calibration point in POD topology Crosstalk is worsen at high calibration point SSN is better at high calibration point BUT noise reduction ratio is smaller Under larger IO noise, jitter can be smaller due to low voltage level

17 Lower Linearity V CC I out V out C P (0~1) High L Low L Low C P (0.35) High C P (0.75)

18 Voltage (V) ISI from Linear Termination High L Low L C P (0~1) Linearity (Ω/V) V LOW (mv) R TERM (Ω) Jitter (UI) 655mV High L Ideal Low C P (0.35) High C P (0.75) mV Low L Low C P (0.35) High C P (0.75) Low linearity has more similar low-voltage (VLOW) to passive termination Under smaller eye-height (-45mV), 13% jitter is improved Time

19 Voltage (V) Crosstalk from Linear Termination High L Low L C P (0~1) Linearity (Ω/V) IO Noise (mv) Jitter (UI) High L Low L Low C P (0.35) High C P (0.75) Low C P (0.35) High C P (0.75) Low linearity has more similar low-voltage (VLOW) to passive termination Under smaller eye-height (-75mV), 12% jitter is improved Time

20 Summary Correct non-linear termination setting is necessary Appropriate calibration point (C P :0~1) At calibration point, non-linear coefficient is calculated Non-linear termination degrades signal quality with DC level shifting Change of crosstalk magnitude Leveling SSN impact on the channel DC shifting from non-linearity is major noise effect in DDR channel

21 Tha k you! --- QUESTIONS?