- 144
- 239 244
Circuit Image
United States
เข้าร่วมเมื่อ 12 ก.ย. 2021
Why Artificial ISI for DFE Design & Verification?
Why Artificial ISI for DFE Design & Verification?
มุมมอง: 106
วีดีโอ
Why Single Pulse Response or Single Shot for ISI Analysis?
มุมมอง 38614 วันที่ผ่านมา
Why Single Pulse Response or Single Shot for ISI Analysis?
Why Half-Rate or Quarter-Rate RX DFE?
มุมมอง 27321 วันที่ผ่านมา
Why Half-Rate or Quarter-Rate RX DFE?
Why Half-Rate or Quarter-Rate Clocking Serializer TX?
มุมมอง 414หลายเดือนก่อน
Why Half-Rate or Quarter-Rate Clocking Serializer TX?
Why Process Evaluation for Transistor, MOSFET Active Device?
มุมมอง 3982 หลายเดือนก่อน
Why Process Evaluation for Transistor, MOSFET Active Device?
Why Process Evaluation for RLC Passive Elements?
มุมมอง 3482 หลายเดือนก่อน
Why Process Evaluation for RLC Passive Elements?
Why Slope Control CTLE in ADC-DSP PAM4 RX?
มุมมอง 5022 หลายเดือนก่อน
Why Slope Control CTLE in ADC-DSP PAM4 RX?
Why Not DFE or Only-1tap Digital DFE in ADC-DSP RX?
มุมมอง 5452 หลายเดือนก่อน
Why Not DFE or Only-1tap Digital DFE in ADC-DSP RX?
Why Dynamic Timing Analysis for Setup & Hold Time?
มุมมอง 4173 หลายเดือนก่อน
Why Dynamic Timing Analysis for Setup & Hold Time?
Why Setup Time & Hold Time Requirements of Latches or Flip-Flops?
มุมมอง 5214 หลายเดือนก่อน
Why Setup Time & Hold Time Requirements of Latches or Flip-Flops?
Why Synchronous or Asynchronous Logic Circuits?
มุมมอง 3044 หลายเดือนก่อน
Why Synchronous or Asynchronous Logic Circuits?
Why Combinational and Sequential Logic Circuits?
มุมมอง 2734 หลายเดือนก่อน
Why Combinational and Sequential Logic Circuits?
Why High-Speed Integrating Mode Phase Interpolator, IMPI?
มุมมอง 5995 หลายเดือนก่อน
Why High-Speed Integrating Mode Phase Interpolator, IMPI?
Why IM PI w/ Inherent 50% Duty Cycle for High-Speed SerDes?
มุมมอง 5735 หลายเดือนก่อน
Why IM PI w/ Inherent 50% Duty Cycle for High-Speed SerDes?
Why An Integrating Mode Phase Interpolator?
มุมมอง 7145 หลายเดือนก่อน
Why An Integrating Mode Phase Interpolator?
Why Not A Voltage Mode or A Current Mode Phase Interpolator?
มุมมอง 8625 หลายเดือนก่อน
Why Not A Voltage Mode or A Current Mode Phase Interpolator?
Why High Linearity Phase Interpolator with Low Power?
มุมมอง 8385 หลายเดือนก่อน
Why High Linearity Phase Interpolator with Low Power?
thanks man!
Hi ToS, you're very welcome. :)
Excellent video thank you for this content
Glad you enjoyed it 🥰
DC Wandering was a very valuable information I was totally unaware of. I can't thank you enough.
Hi OSeyedian, nice to meet you. Thanks you so much for your feedback and I'm glad information was valuable to you. :)
老师可以做一下ADPLL和各型vco的专题吗?
Hi Green tangle, nice to meet you and thanks for the suggestions. I'll do the ADPLL. But I had a few VCOs here: 1. LC VCO Multiphase: th-cam.com/video/J0bKPaAbvxU/w-d-xo.html 2. Feed-forward Delay Cell for A Ring-based VCO: th-cam.com/video/kqL3HRjlums/w-d-xo.html 3. Multiphase VCO for An NRZ or PAM4: th-cam.com/video/lnj6xSTLuOI/w-d-xo.html Please let me know what else you need
thank you!Very well explained
Hi Lab_mangement, thanks for the feedback and I'm glad that you liked it. :)
Hi CC chen, thank you for the amazing tutorial for SERDES. I have one request for you to make some video on how to tune CTLE, REMOTE TX and VGA gains to minimize the ISI . Based on pulse response or frequency response. I tried to use pulse respones as tool to minimize ISI but when I check PRBS simulation result the it do not match with pulse response.
Hi Greywhittef, Thanks for the good suggestions and I'll do it. :) Thanks, CC
!! wow, profesor !!
What are the practical pros and cons of impulse response simulation vs numerically calculating the difference between two step response (a positive and negative step response) simulations?
Hi Joseph, Thanks for the good questions. I have comments below. What are the practical pros and cons of impulse response simulation vs numerically calculating the difference between two step response (a positive and negative step response) simulations? [CC] Are you asking about the correlation between the single-shot align with the two step responses? If yes, my understanding is that it should be equivalent if both responses don't saturate the system, which only assumes the passive elements of the lossy channel. :) Thanks, CC
Thanks for the video. I have a question. The minimum amount of Swing at the Slicer input is decided by "BER/Slicer input noise/Sensitivity/offset" ? What limits the max side? Can we have very High Swing at the Slicer input for both High and low frequency components of the signal? @1:40 : You tell for less dynamic range , sampling accuracy be increased or clockJitter power requirement is reduced? Can you explain in some more detail?
Hi SUMAN, Thanks for the good questions. I have comments below. Thanks for the video. I have a question. The minimum amount of Swing at the Slicer input is decided by "BER/Slicer input noise/Sensitivity/offset" ? What limits the max side? Can we have very High Swing at the Slicer input for both High and low frequency components of the signal? [CC] The maximum swing may distort the linearity of the summing amplifier and then the DFE might not work well. In addition, there's a reliability issue if the maximum swing was not considered in the design. @1:40 : You tell for less dynamic range , sampling accuracy be increased or clockJitter power requirement is reduced? Can you explain in some more detail? [CC] The less dynamic range may help mitigate the sampler's hysteresis and perform faster with a less clock to Q or input setup time. Then the clock requirements of the sampler can be reduced. Thanks, CC
Hi CC, thanks for the sharing, it's really a gread video. I have some queris for the VCO and would like to confirm with you. There are PFD, charge bump and VCO in the PLL system, so when we input the reference clock to PFD, it's doing the phase checking with the feedback frequency, followed by charge pump to convert the phase into voltage, and then input the voltage to VCO, therefore, phase is controlled by the voltage to tune the output frequency. For the VCO, it's input the voltage and output the frequency, please feel free to correct me if I'm wrong. Many thanks
Hi 育瑛,Thanks for the feedback.🙏 Your understanding for the whole PLL loop operation is correct. 😊
@@circuitimage Thanks for the prompt reply!!
Closed eye is very important for measuring super high Speed data analysis 🤔
Hi Lab_mangement, thanks for your feedback. Yes, the eys is is very important for measuring super high Speed data analysis. :)
👍👍😔@@circuitimage
Gorczany Spurs
Hi Christiane, thanks for feedback.
Rath Avenue
Hi Patricia, thanks for the feedback. :)
Beier Spring
Hi Melissa, nice to meet you and thank you for feedback. :)
Blaise Flats
Hi Eric, nice to meet you. :)
Witting Turnpike
Thanks,whatis the rounting complexity of current mode PI
Hi Kobe, thanks for the good question. The big DAC could have lots of parasitic on the high speed routing.
@@circuitimage CC,thanks for your reply,
@@KobeJames-ii4lz You're very welcome. :) Have a nice weekend!!
I m afraid it s the opposite around. One should reduce the capacitance.
Hi @arnaud.lancelot, thanks for the feedback. Could you please elaborate on your concern?
@@circuitimage If the diff pair acts like an integrator, the gain has the form of gm/C. The input noise is (kt/C) /(gain x gain) . So input noise should look like kTxC/(gm x gm). So boost the gm transconductance for sure but diminish the capacitance C.
@@arnaud.lancelot I agree. So, in my video, I mentioned, the bigger C the less noise, but the speed goes down.
@@circuitimage What i m saying is that you should look at the SNR, signal to noise ratio. The smaller the capacitance, the lower the noise from... the input perspective. So you increase the SNR at the input. But you also increase it at the output (gmxgmxVxV/CxC)/(kT/C) ~1/C So you reduce C. you go faster and has a better SNR. Reducing the noise is useless if you reduce even more the signal gain.
@@arnaud.lancelot Thanks for the feedback, 🙏 and I agree with your SNR, which should match the input referred noise, which is the ouptut noise/gain, not just noise itself. 🎉😊
Why do we need equalization
Hi Ahmad, Nice to meet you and thank you 🙏 for your questions. Since the channel frequency response was not equal loss over frequency, so we need to equalize it. 😊
@@circuitimage channel frequency response loss didn't equal the which frequency?
@@NostalgiaT Ideally, you want the gain is the same over your interested frequency, which is flat. But a lossy channel isn't flat.
@@NostalgiaT This CTLE th-cam.com/video/zsuJMqadaKY/w-d-xo.htmlsi=vpxIC-vZPHHTYV3d may show it.
What's the difference between ctle and dfe?
Hi Ahmad, Nice to meet you and thank you 🙏 for your questions. Lots difference and you might want to check all my videos😊
Jacobson Walks
Thank you!!
Thanks for the nice video! Is 3D packaging the same as 'chiplet'? Looking forward to more videos on packaging!
Hi Hanyue, Nice to meet you. Thanks for the good suggestion. Yes, that's correct. The 3D packaging helps the implementations of the Chiplet. Thanks, CC
Great videos on channel !! But one doubt is how can the nyquist rate be half of data rate....shouldn't it be twice? i.e. 20Ghz and not 5Ghz
Hi Shivangi, Nice to meet you. Thanks for the good feedback. The 5GHz would be the Nyquist rate of 10Gbps. Thanks, CC
@@circuitimage Thanks for your response... But it would be quite helpful if you can explain this..I am a bit confused
@@betu1207 Hi Shivangi, Sure. The simplest way of thinking about this is taking the maximum transition from any data pattern is the clock pattern, which is 1010. For example, the 1UI of the 10Gbps is 100ps, but the 1010 or clock pattern is 5GHz since the clock period is 200ps instead of 100ps or 50ps, so 5GHz is the Nyquist rate of the 10Gbps.
Hi,CC,could you introduce me the "background calibration"?thanks!
Hi Kobe, sure. Nice to meet you. :) Is this ( th-cam.com/video/YsFg_ReBBzQ/w-d-xo.html ) you're looking for?
Nice video as always CC and pointing out advantages of FR DFE. With data rates going up and difficulty for PLL to generate FR clocks for Rx, HR clocking is becoming popular. E.g. for PCI-Gen5 32Gsps, I still see most PLL outputting 16GHz clock, so DFE then has to be HR.
Hi Abhirup, Thank you 🙏 so much for your feedback, which I agree with a lot. My PCI-Gen5 32Gbps also applied PLL's 16GHz clock.😊 I'll put your comments in my next video: "Why half-rate or quarter-rate DFE." 😊
@@circuitimage great, look forward CC.
@@abhiruplahiri1 Thanks for your always support. :)
The interface data rate can reach up to 100Gb/s per lane. If we use a full-rate decision feedback equalizer (DFE), generating a clock signal becomes challenging and requires significant power. Thus, achieving 200Gb/s could be difficult due to the complexity of generating a full-rate clock at such high speeds.
Hi 장구, Nice to meet you. Thank you 🙏 so much for your feedback, which I agree with a lot. I'll put your comments in my next video: "Why half-rate or quarter-rate DFE." 😊
Great content as always. I did note that slide 4 title has a typo. The slide title has "...Issue at the of the..." I assume you were debating to use "at" or "of". I'd suggest "of" as the phenomenon is not the location of the summers, but rather a property of having two summer instances.
Hi Matthew, Nice to meet you. Thank you 🙏 so much for your corrections and for looking carefully, accidentally adding extra "the of".
Hi ,can you suggest some tap implementation techniques for DFE.
Hi Pran, nice to meet you and thank you for the suggestions. Is this (th-cam.com/video/BZLVH-8JFys/w-d-xo.htmlsi=vhs3nMSafdXXshHz) you're looking for?
Why is it said that a coefficient greater than or less than 0 means there is no requirement?
Hi Yifeng, thanks for the question. Could you please elaborate on which time your refer to?
Thank you😊
Hi Lab_mangement, you're very welcome. :)
Thanks for sharing!
Hi PeiLife, you're very welcome.
Hello, why need to add SR latch after strong arm latch? Is it to convert analog signals into logical signals?😊
Hi 树多多, you can check my video: Why Strong-arm Latch w/ An Optimized RS Latch to Meet Digital Timing? th-cam.com/video/vMWY41DADFc/w-d-xo.htmlsi=R2mkHLYYmrRiC9P3
@@circuitimage thanks!
@@树多多 ur very welcome. :)🥰
Brock Squares
Klocko Row
Thank you so much.
Hi, CC, Thanks for keeping posting new videos! I have a questions. at time 12.14 when you talked about the duty-cycle impact, I didn't fully understand why you said full-rate clock have accurate 50% duty cycle while in the half rate case the duty cycle can cause problem. Could you elaborate a bit on that? Thanks!
I guess the point is duty cycle in FR sampling can eventually be sensed as an error and corrected for by the loop. Such impairment worsens the ber even in FR, but in HR the duty cycle error on half rate clocks may not even be sensed (eg A goes to left, B to right but T is aligned) leading to higher uncorrected error and thus higher ber. I let CC comment more and enlighten us.
Hi Xiaodong, Thank you so much for the good questions. Also, thanks for the Abhirup's very good feedback, which aligns with what I tried to say. The duty-cycle distortion (DCD) may have a BER impact on all CDR no matter which topology or clock rate; therefore, we must do the duty-cycle correction (DCC) in another loop. The FR sampling's data & transition sample usually has a better DCC than the HR; therefore, the DCD of the FR is usually less. But adding a good design of DCC in the HR still can minimize the DCD as much as possible.
Hi Abhirup, thanks for your very good feedback, which aligns with what I tried to say. :)
@@circuitimage many thanks CC. Always looking forward to your interesting videos.
Great.....can you please explain a bit more on how you choose the coefficient values of Pre and Post cursor taps. And 2nd doubt is can we see any case where you keep both the coefficient to be positive and in that case if the transition parts will have higher swing or lower ?
Hi Surya, Thank you so much for the good questions. I have comments below. Great.....can you please explain a bit more on how you choose the coefficient values of Pre and Post cursor taps. [CC] I did the a single-pulse response to figure out the coefficients of Pre and Post cursor, respectively. And 2nd doubt is can we see any case where you keep both the coefficient to be positive and in that case if the transition parts will have higher swing or lower ? [CC] Yes, that's possible. We only focus on the data parts, but the transition parts could have a higher swing or lower due to the reflection. Thanks, CC
Sir you just saved my ass again
Hi Ryan, I'm glad this helped you again. :)
Hi Dr. Chen, thank you so much for your fruitful videos! I have two questions: 1. Are there any recent products of TSPC FF being used in the industry? 2. Samsung presented a new TSPC FF design in the VLSI symposium 2024, “An Area-Efficient True Single-Phase Clocked and Conditional Capture Flip-Flop for Ultra-Low-Power Operations in 7nm Fin-FET Process” but it seems like a static FF instead of a dynamic one. From the viewpoint of ultra-low voltage operation capability(~0.2V), energy consumption, and robustness, which is preferable?
Hi XX, Nice to meet you and you're very welcome. I have comments below. have two questions: 1. Are there any recent products of TSPC FF being used in the industry? [CC] Yes, that's a lot. 2. Samsung presented a new TSPC FF design in the VLSI symposium 2024, “An Area-Efficient True Single-Phase Clocked and Conditional Capture Flip-Flop for Ultra-Low-Power Operations in 7nm Fin-FET Process” but it seems like a static FF instead of a dynamic one. From the viewpoint of ultra-low voltage operation capability(~0.2V), energy consumption, and robustness, which is preferable? [CC] Unfortunately, IEEE has not put in online yet and I cannot check the paper. Would you mind sending it to me or share in your sharepoint? Thanks, CC
@@circuitimage Thank you so much for your reply. Here is the link to the paper: vlsi24.mapyourshow.com/mys_shared/vlsi24/handouts/C25.2_Thu_Hwang.pdf
@@circuitimage Thank you very much for your reply! I can't send you the Google Docs sharing link successfully. I guess the comments with links are automatically blocked in your channel settings. You can check whether Comment->Community->Automated Filters->Block Links is checked.
the paper link: drive.google.com/file/d/1kWNMA77yZmEInZci1twO1B7JAaringy-/view?usp=drive_link
@xxw6371 Thank you so much for the paper, which is very promising. But I'm not sure if that's really robust enough. 1. The input is not single-ended as the traditional TSPC 2. The differential signaling is good for PSRR, which might not be needed, but it adds the mismatch concerns. 3. Lots of elements of their LCFF may have more leakage and lots of parasitic than the traditional TSPC
From you kind sharing, it seem to me the reason we downgrade from full-rate to half-rate and to quarter-rate clocking is mainly due to speed limitation. If speed is no concern, we always choose full-rate clocking? I would also like to know from power perspective, full-rate, half-rate and quarter-rate, which one is more power efficient? Any other advantage for quarter-rate clocking? Thanks.
Hi MrNonamel, nice to meet you and thanks for the good question. 😊 From power perspective, quarter-rate clocking is lower than the half-rate, and the half-rate is lower than the full-rate. But, the assumption is those phase correction or DCC and the complicated multi-phase routing is manageable in layout and design. :)
@@circuitimage DDR5 DRAM divide CK clock by 2 to achieve quarter-rate clocking, I wonder if this is due to speed or power reason?
Hi MrNonamel, Thank you so much for the good questions. I have comments below. DDR5 DRAM divide CK clock by 2 to achieve quarter-rate clocking, I wonder if this is due to speed or power reason? [CC] Both 😉 In addtion, the quarter-rate clocking in DDR5 DRAM would reduce the EMI (Electromagnetic Interference) challenges & SI issue, which also related to cost. 😊 Thanks, CC
Thanks CC for the informative video. One more thing to add as a limitation of full rate clocking especially as speeds go up is the design of PLL VCO at full-rate. It is easier to make a more phase noise optimized VCO at half rate, e.g. 16GHz for a 32gsps link.
Hi Abhirup, thanks for the reminder. I'm kind of implying it at the last Why half rate SerDes and this Serilaizer TX (if good RJ/DJ from the full-rate is doable in slide #2). :)
@@circuitimagemany thanks CC. Kudos to you for sharing such great information.
@@abhiruplahiri1 Thank you so much as well 🙏for your valuable feedback, which helps me improve the shared info and benefits other people in this IC design society. 😉
Hello, thanks you very much for your videos. I have one question about minimum broadband noise described in the spec, is it differential rms nosie or rxm rmp ports each have uncorrelated 5.2mV RMS noise?
Hi Gman, nice to meet you. Good question. It's differential uncorrelated 5.2mV RMS noise. :)
Thank you very much! Can you do a video about Frequency divider using CML Flip Flop ? Thanks in advanced!
Hi Viet, nice to meet you and thank you for your suggestions. I'll do a video about Frequency divider using CML Flip Flop. :)
How is it possible to have a architecture without drawback? Look forward to the following video about half-rate or quad-rate mode.
Hi Analogue, nice to meet you, and thanks for the feedback. Yes, we'll also show the drawback of the half-rate or quad-rate clocking later. :)
really good video! Helped me alot
Hi Syed, nice to see you and thank you for the feedback. I'm happy it helped you. :)
You are the best.
Hi layt01, nice to meet you and thank you for your feedback. I'm glad you liked it. :)
That's a nice video CC. We can do one octave tuned PLL (1:2), we can possibly generate any lower freq. by div 2, 4...etc. The challenge of course is to do a 1:2 LC VCO with good phase noise. May be food for thought for another video 😀 For lower speed multi-protocol use cases, may be ring VCO is already good enough, any inputs?
Hi Abhirup, thanks for the nice feedback. Your experiences match what I have. My understanding is the NRZ up to 32Gbps, the ring VCO is already good enough, but not good enough for the PAM4 SerDes above 56Gbps. Therefore, the :2 LC VCO with good phase noise is Challenge. :)
@@circuitimage thanks for the input CC.
@@abhiruplahiri1 you're very welcome. :)
this is a talented youtube channel I found for analog mixed signal field!
Hi Zixiang, thanks for the feedback and nice to meet you. I'm glad you liked it. :)
Very interesting Serdes tech 😁
HI Allen, thanks for the feedback.
謝謝您! 很棒耶!👍
Hi Layout_books, 非常感谢🙏😉
un saludo
Hi HubbA, Thank you.
@@circuitimage I encourage you to continue, but what I would like is to solve problems, but from the basics, you. It has very advanced concepts, it should be created as a course, in which it would give the possibility of getting at their level, VD. MUST teach (teacher :), ! I follow him :), greetings from Spain
@@k4r4m310. Hi HubbA, Thank you for your encouragement and suggestions. I'll keep going and try to make a course for everyone. Likewise, I may think about writing a book for everyone if time is allowed. :)
@@circuitimage do it yourself , hágalo vd. un cordial saludo